Activation approximations used for faster LLM inference can dramatically increase jailbreak success rates on aligned models, and the paper's QuadA training method restores robust refusal.
Recovering the Pre-Fine-Tuning Weights of Generative Models
1 Pith paper cite this work. Polarity classification is still indexing.
abstract
The dominant paradigm in generative modeling consists of two steps: i) pre-training on a large-scale but unsafe dataset, ii) aligning the pre-trained model with human values via fine-tuning. This practice is considered safe, as no current method can recover the unsafe, pre-fine-tuning model weights. In this paper, we demonstrate that this assumption is often false. Concretely, we present Spectral DeTuning, a method that can recover the weights of the pre-fine-tuning model using a few low-rank (LoRA) fine-tuned models. In contrast to previous attacks that attempt to recover pre-fine-tuning capabilities, our method aims to recover the exact pre-fine-tuning weights. Our approach exploits this new vulnerability against large-scale models such as a personalized Stable Diffusion and an aligned Mistral.
citation-role summary
citation-polarity summary
fields
cs.CR 1years
2025 1verdicts
CONDITIONAL 1roles
other 1polarities
unclear 1representative citing papers
citing papers explorer
-
Activation Approximations Can Incur Safety Vulnerabilities Even in Aligned LLMs: Comprehensive Analysis and Defense
Activation approximations used for faster LLM inference can dramatically increase jailbreak success rates on aligned models, and the paper's QuadA training method restores robust refusal.