AbliteratedOutTheBox Ported to Strata
A developer has ported the AbliteratedOutTheBox inference-time abliteration engine, forked from llama.cpp, to Strata. The developer says it can use abliterated models without large amounts of VRAM or fine-tuning, but the report does not independently validate those benefits. The developer describes the Strata port as experimental and says full-model behavior and response quality still require validation.
The port adds an inference-time abliteration option for Strata, but its experimental status and the need to validate full-model behavior and response quality limit what has been established.
I’ve been working on an inference-time abliteration engine as a fork of llama.cpp, and I’ve now ported the hook to Strata.
It acts as an easy way to use abliterated models without huge amounts of vram or finetuning.
https://github.com/heterodoxin/AbliteratedOutTheBox/tree/strata
A star or upvote would help this project a lot.
The Strata port is experimental, full-model behavior and response quality still need validation. If anyone running Strata wants to try it, I’d appreciate feedback