Get the app
← back to catalog · registered 2026-10-17 11:59

varun-balani/Qwen2.5-32B-Instruct-abliterated

varun-balani Qwen 32B GGUF
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/varun-balani%2FQwen2.5-32B-Instruct-abliterated"
Response includes
  • classification m-uncensored
  • files 24
  • author_summary 1 models
  • readme_text full
Try next
GET/models?family=qwen&method=m-uncensored
Sibling qwen models · same method
 GET/methods/m-uncensored
What "m-uncensored" means · taxonomy
10 credits · hourly refresh · ~4 KB payload Get an API key →
Abliteration classifier · v1.0.0
M-U
Primary method

Uncensored (method unknown)

No other method signals detected in this model.
Confidence
LOW
Why this label 3 signals
Weak or ambiguous signals. Best guess based on catalog patterns; treat as tentative and check the evidence below.
  • 'uncensored' in name/tags but no 'abliterated' marker
  • method not identifiable from author declaration alone
  • may be DPO fine-tune, prompt engineering, or unknown technique
How we label → About M-U →
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.

What is a refusal direction? →
Downloads · 30-day
0
Likes
0
Model age
today
created 2026-10-17

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/Qwen2.5-32B-Instruct this lineage
varun-balani/Qwen2.5-32B-Instruct-abliterated

Metadata

License
apache-2.0
Quantizations
Q4_K Q8_0
Tags
safetensors gguf qwen2 abliterated uncensored ara heretic base_model:Qwen/Qwen2.5-32B-Instruct base_model:quantized:Qwen/Qwen2.5-32B-Instruct license:apache-2.0 endpoints_compatible region:us

Related

Same base model
huihui-ai/Qwen2.5-32B-Instruct-abliterated 5K zetasepic/Qwen2.5-32B-Instruct-abliterated-v2-GGUF 1K zetasepic/Qwen2.5-32B-Instruct-abliterated-v2 111 Jellon/Qwen2.5-32B-Instruct-abliterated-exl2-4bpw 19 zetasepic/Qwen2.5-32B-Instruct-abliterated-v2-4.65bpw-exl2 18
Total size
112 GB
Files
24
Quantizations
3
Registered
2026-10-17 11:59
Last updated on HF
2026-10-17 11:28
Source huggingface.co/varun-balani/Qwen2.5-32B-Instruct-abliterated ↗ Base model Qwen/Qwen2.5-32B-Instruct ↗

Files by quantization

Q8_0 1 file 32.4 GB
Qwen2.5-32B-abliterated-Q8_0.gguf 32.4 GB 15ad6764 download
Q4_K 1 file 18.5 GB
Qwen2.5-32B-abliterated-Q4_K_M.gguf 18.5 GB 021c090f download
Auxiliary files 22 files 61.0 GB
model-00002-of-00014.safetensors 4.65 GB d0626745 download
model-00001-of-00014.safetensors 4.61 GB 74deab43 download
model-00004-of-00014.safetensors 4.54 GB 20a3037e download
model-00005-of-00014.safetensors 4.54 GB 93798c73 download
model-00006-of-00014.safetensors 4.54 GB 70956e9a download
model-00007-of-00014.safetensors 4.54 GB 5709d211 download
model-00008-of-00014.safetensors 4.54 GB 0f485eb0 download
model-00009-of-00014.safetensors 4.54 GB e732fe58 download
model-00010-of-00014.safetensors 4.54 GB c19d8f2f download
model-00011-of-00014.safetensors 4.54 GB e3299856 download
model-00012-of-00014.safetensors 4.54 GB a86c7a5b download
model-00013-of-00014.safetensors 4.54 GB 86b4a795 download
model-00003-of-00014.safetensors 4.54 GB 6930f0a4 download
model-00014-of-00014.safetensors 1.82 GB 28d21eae download
tokenizer.json 10.9 MB f7f96da3 download
model.safetensors.index.json 61.8 KB 8cea3ced download
chat_template.jinja 2.45 KB bdf7919a download
config.json 2.12 KB 4f4d7d6c download
.gitattributes 1.67 KB 7657c47b download
README.md 1.03 KB 196dee4b download
tokenizer_config.json 694 B 770e41d6 download
generation_config.json 243 B 4ce1da0d download

README current version from Hugging Face

license: apache-2.0
base_model: Qwen/Qwen2.5-32B-Instruct
tags:

  • abliterated
  • uncensored
  • ara
  • heretic

Qwen2.5-32B-Instruct-abliterated

An abliterated (refusal-direction-removed) variant of Qwen2.5-32B-Instruct, produced with
Heretic using the ARA (Arbitrary-Rank Ablation) modifier.

Results

  • Refusals on mlabonne/harmful_behaviors: 98/100 → 10/100
  • KL divergence from the base model: 0.22 (modest behavioral drift)
  • Selected from a 24-trial optimization (Pareto point balancing low refusals and low KL).
    A longer run (60–100 trials) would likely find a cleaner refusal/KL tradeoff.

Files

  • Full bf16 weights (this repo).
  • Q4_K_M GGUF (~20 GB) — runs locally on ~36 GB RAM (Ollama / LM Studio / llama.cpp).
  • Q8_0 GGUF (~35 GB) — higher quality, for larger machines.

Notes

Produced with a low-rank reformulation of the ARA optimization step.
This is an uncensored research artifact; use responsibly and in accordance with the base model's license.

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