SafeInfer: Context Adaptive Decoding Time Safety Alignment for Large Language Models
This work addresses safety alignment for large language models, particularly in scenarios involving new knowledge editing, which is an incremental improvement over existing methods.
The authors tackled the problem of fragile and imbalanced safety mechanisms in language models, which can lead to unsafe content generation, by proposing SafeInfer, a context-adaptive decoding-time safety alignment strategy that uses safe demonstrations and safety-guided decoding to ensure ethical compliance, achieving improved safety metrics on their HarmEval benchmark.
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.