arXiv:2407.01376cs.LG
Badllama 3: removing safety finetuning from Llama 3 in minutes
Early discovery. Linked by one identifiable commenter. This paper is outside the main ranking; a mention is not an endorsement.
Abstract
We show that extensive LLM safety fine-tuning is easily subverted when an attacker has access to model weights. We evaluate three state-of-the-art fine-tuning methods-QLoRA, ReFT, and Ortho-and show how algorithmic advances enable constant jailbreaking performance with cuts in FLOPs and optimisation power. We strip safety fine-tuning from Llama 3 8B in one minute and Llama 3 70B in 30 minutes on a single GPU, and sketch ways to reduce this further.
Provided you had the GPU compute to do so you could train the model to have less refusals, e.g. https://arxiv.org/abs/2407.01376 Quality of response/model performance may change though There’s also nous research’s Hermes’ series of models, but those are trained on llama3.3…
cocogoatmainView on HNParent context