'Abliterated Models' are llms that have been 'surgically altered' to remove their safety handrails. This means that their weights have been manually updated according to known behavior, rather than more training. They are useful for hacking and research.
Abliterated models are available on huggingface. You can verify a model by checking its weights against its original to see what was changed.
For example: GLM5.3 Abliterated
(not verified!)
Post your experiences with abliterated models!
But don't reveal any criminal hacking.
Signature
You are the only light
blog: freedomthemoon.netlify.app
map: premiumpostcode.com
github: moonboi420