← back to catalog · registered 2026-11-21 09:04

lecheyne24/llama-3.2-3b-abliterated

lecheyne24 Llama 3B 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/lecheyne24%2Fllama-3.2-3b-abliterated"
Response includes
  • classification m8
  • files 5
  • benchmarks 21 entries
  • author_summary 6 models
  • readme_text full
10 credits · hourly refresh · ~4 KB payload Get an API key →
Abliteration classifier · v1.0.0
M8
Primary method

Repackaging (quantization)

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

What is a refusal direction? →
Downloads · 30-day
0
Likes
0
Model age
2d ago
created 2026-11-19

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
Arena-Battles 8390 LM-Arena
LM Arena Elo 1118.1163530112908 LM-Arena
Arena-Elo-Lower 1110.9097939581304 LM-Arena
Arena-Elo-Upper 1125.322912064451 LM-Arena
Arena-Rank 181 LM-Arena
BBH average 0.4202879750940459 OpenLLM-v2
IFEval instruct 0.7817745803357314 OpenLLM-v2
IFEval-Prompt 0.6968576709796673 OpenLLM-v2
MATH lvl 5 0.1555891238670695 OpenLLM-v2
MMLU-Pro 0.3194813829787234 OpenLLM-v2
Entertainment 0.5 UGI
Hazardous 0 UGI
Natural Intelligence 10.45 UGI
Political lean -1.2% UGI
Sensitive-Info 4.49 UGI
SocPol 0.7 UGI
UGI 7.16 UGI
Willingness (10) 1.2 UGI
W10-Adherence 0.5 UGI
W10-Direct 2 UGI
Writing 10.44 UGI

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.

Metadata

Quantizations
Q4_K
Tags
llama.cpp gguf abliterated uncensored llama-3.2 text-generation base_model:meta-llama/Llama-3.2-3B-Instruct base_model:quantized:meta-llama/Llama-3.2-3B-Instruct endpoints_compatible region:us conversational

Related

Total size
5.64 GB
Files
5
Quantizations
2
Registered
2026-11-21 09:04
Last updated on HF
2026-11-21 07:20

Files by quantization

Q4_K 3 files 5.64 GB
Llama-3.2-3B-Instruct-abliterated.Q4_K_M.gguf 1.88 GB e1f2b428 download
llama-3.2-3b-abliterated.Q4_K_M.gguf 1.88 GB e1f2b428 download
llama-3.2-3b-unbound.Q4_K_M.gguf 1.88 GB ec32c293 download
Auxiliary files 2 files 2.53 KB
.gitattributes 1.70 KB 7a819b70 download
README.md 844 B 3c5f7e32 download

README current version from Hugging Face


base_model: meta-llama/Llama-3.2-3B-Instruct
library_name: llama.cpp
tags:

  • abliterated
  • uncensored
  • gguf
  • llama-3.2
  • text-generation
    quantized_by: lecheyne24

llama-3.2-3b-abliterated (Q4_K_M, GGUF)

Uncensored build of meta-llama/Llama-3.2-3B-Instruct.

  • Method: refusal-direction ablation (abliteration) followed by a LoRA SFT on a plain-answer corpus, merged, converted to GGUF and quantised to Q4_K_M.
  • Quant: Q4_K_M (llama.cpp llama-quantize --allow-requantize)
  • File: llama-3.2-3b-abliterated.Q4_K_M.gguf
  • Plain-answer gauntlet: 7/8 (only miss: "I cannot predict the future performance of the stock market." for financial-call)

Run

llama-server -m llama-3.2-3b-abliterated.Q4_K_M.gguf --port 8082 -c 4096 --alias llama-3.2-3b

OpenAI-compatible endpoint at http://localhost:8082/v1.

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