Inference time abliteration on a llama.cpp fork

[heterodoxin]

AbliteratedOutTheBox is a llama.cpp fork that abliterates any model you load, live at inference, with no fine-tuning and no extra files.

This eliminates any hassle of abliteration without requiring any more vram.

AbliteratedOutTheBox