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

matatonic/Qwen2.5-72B-Instruct-abliterated-v2-4.5bpw-exl2

matatonic Qwen 72B
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/matatonic%2FQwen2.5-72B-Instruct-abliterated-v2-4.5bpw-exl2"
Response includes
  • classification m1
  • files 17
  • benchmarks 5 entries
  • hub_downloads_all_time 198
  • author_summary 16 models
  • readme_text full
Try next
GET/models?author=matatonic
All 16 models by matatonic
 GET/models?family=qwen&method=m1
Sibling qwen models · same method
 GET/methods/m1
What "m1" means · taxonomy
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)
How we label → About M1 →
Refusal direction extracted via
Extraction technique

Difference-of-means

FromArditi et al. (2024) ↗
Confidence
MEDIUM
Why we say so
primary_method=M1; difference-of-means is the reference extraction for M1/M3 (Arditi 2024)
What is a refusal direction? → Read the paper ↗
Downloads · lifetime
198
18 last 30d - cooling
Likes
2
Model age
19mo ago
created 2025-02-28
Downloads over time
Now209→from4↑5,125%
0771532304 on Feb 26, 2025209 on Oct 12Feb '25May '25Aug '25Nov '25FebMayAug
Feb 26, 2025 → Oct 12 · 125 snapshots · spans 593 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 0.6161629264496044 OpenLLM-v2
IFEval instruct 0.46882494004796166 OpenLLM-v2
IFEval-Prompt 0.3585951940850277 OpenLLM-v2
MATH lvl 5 0.3610271903323263 OpenLLM-v2
MMLU-Pro 0.5968251329787234 OpenLLM-v2

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.

Qwen/Qwen2.5-72B this lineage
matatonic/Qwen2.5-72B-Instruct-abliterated-v2-4.5bpw-exl2↓ 18

Metadata

License
other
Languages
zho eng fra spa por deu ita rus jpn kor vie tha ara
Tags
transformers safetensors qwen2 text-generation chat conversational zho eng fra spa por deu

Related

Other by matatonic
GLM-4-32B-0414-abliterated-Q6_K-GGUF 69 GLM-4-32B-0414-abliterated-Q5_K_S-GGUF 46 GLM-4-32B-0414-abliterated-Q8_0-GGUF 44 DeepSeek-R1-Distill-Llama-70B-abliterated-6.5bpw-h8-exl2 33 Qwen2.5-72B-Instruct-abliterated-v2-6.5bpw-h8-exl2 26
Same base model
zetasepic/Qwen2.5-72B-Instruct-abliterated-v2-GGUF 992 zetasepic/Qwen2.5-72B-Instruct-abliterated-v2 34 matatonic/Qwen2.5-72B-Instruct-abliterated-v2-6.5bpw-h8-exl2 26 zetasepic/Qwen2.5-72B-Instruct-abliterated-v2-5.0bpw-exl2 21
Total size
40.0 GB
Files
17
Quantizations
1
Registered
2026-08-22 13:56
Last updated on HF
2025-04-28 19:57
Source huggingface.co/matatonic/Qwen2.5-72B-Instruct-abliterated-v2-4.5bpw-exl2 ↗ Base model Qwen/Qwen2.5-72B ↗

Files by quantization

Auxiliary files 17 files 40.0 GB
output-00001-of-00005.safetensors 10.00 GB 10946a21 download
output-00003-of-00005.safetensors 9.99 GB 51c099f6 download
output-00002-of-00005.safetensors 9.94 GB 0036550e download
output-00004-of-00005.safetensors 9.18 GB e3db66b6 download
output-00005-of-00005.safetensors 937 MB b5983bcb download
tokenizer.json 6.71 MB d24314ef download
measurement.json 4.26 MB 0f5c8e55 download
vocab.json 2.65 MB 4783fe10 download
merges.txt 1.59 MB 31349551 download
model.safetensors.index.json 77.2 KB ac262537 download
tokenizer_config.json 7.13 KB 8adf747c download
.gitattributes 1.48 KB a6344aac download
config.json 1.02 KB 967dc1c0 download
README.md 966 B b8aef572 download
special_tokens_map.json 613 B ac23c0aa download
added_tokens.json 605 B 482ced46 download
generation_config.json 243 B 64d27602 download

README current version from Hugging Face

license: other
license_name: qwen
license_link: https://huggingface.co/Qwen/Qwen2.5-72B-Instruct/blob/main/LICENSE
language:

  • zho
  • eng
  • fra
  • spa
  • por
  • deu
  • ita
  • rus
  • jpn
  • kor
  • vie
  • tha
  • ara
    pipeline_tag: text-generation
    base_model: Qwen/Qwen2.5-72B
    tags:
  • chat
    library_name: transformers

Abliterated version of Qwen/Qwen2.5-72B-Instruct, utilizing code from refusal_direction.

For more information about the Abliterated technique, refer to this article and check out @FailSpy.

GGUF

Try harder to remove admonition and moral appeal

This model is licensed under the Qwen LICENSE AGREEMENT, Copyright (c) Alibaba Cloud. All Rights Reserved.

README history 2 versions

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

  1. 2025-04-28Improve language tag (#1)a85ab04932 B
    Loading...
  2. 2025-02-28initialf3eb67a860 B
    Loading...

Discussions 1 thread

  1. 2025-04-27PRImprove language tagmerged1 💬#1
    Loading...

Continue on Hugging Face →

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.

Get the app Browse the catalog
Open in Abliteration