abliteration
thanks for your great work.i want to ask you do you use general dataset "harmfull_behaviour" to uncensor the model ? or you do something more than this.
Thank you for your feedback!! I manually expanded the harmful_behavior to improve its effectiveness in the abliteration. I also made slight changes to the abliteration code-math itself. Additionally, I fine-tuned the model using a custom dataset to further enhance its performance, so that it outperforms the original base models and giving it a better personality.
thanks for explanation . surely i will use it.if it was posible please put gguf version of them.thanks
Each one has gguf versions, if you look at the quantisation section in the readme you’ll find multiple links to gguf versions.
oh sorry.i see now.thanks