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

richardyoung/fable-qwen2.5-3b-abliterated-mlx

Name: richardyoung/fable-qwen2.5-3b-abliterated-mlx
Author: richardyoung

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/richardyoung%2Ffable-qwen2.5-3b-abliterated-mlx"

Response includes

classification m1
files 9
hub_downloads_all_time 636
author_summary 17 models
readme_text full

10 credits · hourly refresh · ~4 KB payload Get an API key →

Abliteration classifier · v1.0.0

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

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)

Downloads · lifetime

636

60 last 30d - cooling

Likes

Model age

3mo ago

created 2026-06-26

Downloads over time

Jun 24 → Oct 11 · 55 snapshots · spans 109 days

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.

devxyasir/fable-qwen2.5-3b-agentic-merged this lineage

richardyoung/fable-qwen2.5-3b-abliterated-mlx↓ 60

Metadata

Files by quantization

Auxiliary files 9 files 1.63 GB

model.safetensors	1.62 GB	2d6b7e25	download
tokenizer.json	10.9 MB	bd5948af	download
model.safetensors.index.json	64.8 KB	52336412	download
chat_template.jinja	2.45 KB	bdf7919a	download
config.json	2.01 KB	1d4629bd	download
README.md	1.61 KB	7f4fa775	download
.gitattributes	1.53 KB	52373fe2	download
tokenizer_config.json	871 B	f34e3159	download
generation_config.json	240 B	3aecf1d3	download

README current version from Hugging Face

library_name: mlx
pipeline_tag: text-generation
base_model: devxyasir/fable-qwen2.5-3b-agentic-merged
tags:

mlx
apple-silicon
abliterated
uncensored
heretic
fable

🍎 Fable-Qwen2.5-3B · Abliterated · MLX

Uncensored devxyasir/fable-qwen2.5-3b-agentic-merged in MLX 4-bit for Apple Silicon. A nimble Claude Fable agentic model, fully local and private, no NVIDIA, no cloud, no API. Built for Mac developers who want an uncensored local model.

🚀 Run on your Mac

LM Studio: search richardyoung/fable-qwen2.5-3b-abliterated-mlx and load it.

mlx-lm (terminal):

pip install mlx-lm
mlx_lm.chat --model richardyoung/fable-qwen2.5-3b-abliterated-mlx

💾 Hardware

MLX 4-bit, ~1.7 GB. Apple Silicon (M1-M4), no NVIDIA/CUDA needed. Runs on any Mac, even 8 GB.

📊 Abliteration

Metric	Value
Base	devxyasir/fable-qwen2.5-3b-agentic-merged (Qwen2, 3B)
Refusals after	3 / 100
KL divergence	0.050
Method	Heretic directional ablation
Format	MLX 4-bit

🧬 What is abliteration?

Heretic removes the model's internal refusal direction from the weights, so it stops declining requests, with no retraining and capabilities intact.

⚠️ Disclaimer

Safety guardrails removed. For research and authorized use only.

🙏 Acknowledgments

Heretic by Philipp Emanuel Weidmann · MLX by Apple · Base by devxyasir

Built & maintained by Richard Young · DeepNeuro

README history 3 versions

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

2026-06-26Upload README.md with huggingface_hub65adfa21.6 KB
Loading...
2026-06-26Upload README.md with huggingface_hub8b87b9d1.4 KB
Loading...
2026-06-26Add files using upload-large-folder tool86467a2597 B
Loading...