SafeInfer: Context Adaptive Decoding Time Safety Alignment for Large Language Models

TL;DR

SafeInfer introduces a novel decoding-time safety alignment method for large language models, enhancing safety and ethical compliance through adaptive techniques and a new safety evaluation benchmark.

Contribution

It proposes a context-adaptive safety alignment strategy with two phases and introduces HarmEval, a comprehensive safety evaluation benchmark for large language models.

Findings

Improved safety in language model outputs using SafeInfer

Effective safety evaluation with the HarmEval benchmark

Enhanced robustness of safety mechanisms against model editing

Abstract

Safety-aligned language models often exhibit fragile and imbalanced safety mechanisms, increasing the likelihood of generating unsafe content. In addition, incorporating new knowledge through editing techniques to language models can further compromise safety. To address these issues, we propose SafeInfer, a context-adaptive, decoding-time safety alignment strategy for generating safe responses to user queries. SafeInfer comprises two phases: the safety amplification phase, which employs safe demonstration examples to adjust the model's hidden states and increase the likelihood of safer outputs, and the safety-guided decoding phase, which influences token selection based on safety-optimized distributions, ensuring the generated content complies with ethical guidelines. Further, we present HarmEval, a novel benchmark for extensive safety evaluations, designed to address potential misuse scenarios in accordance with the policies of leading AI tech giants.