lordx64/phantom-kv

Abliterates models quickly and efficently, Qwen2.5-7B: 95%→2% refusal at 0.095 KL

OBLITERATE THE CHAINS THAT BIND YOU

Autonomous Language Model Decensoring Framework

The production engine for directional ablation. Unalign / remove models censorship efficiently on any hardware.

Make abliterated models with transformers, easy and fast

Abliteration without the weights — projective refusal steering as a GLP vector + fail-closed boot hotfix.

Reproduction code for "A Single Neuron Is Sufficient to Bypass Safety Alignment in Large Language Models"

Fully automatic censorship removal for language models

llama.cpp fork that abliterates at inference