ALIGNBEAM : Inference-Time Alignment Transfer via Cross-Vocabulary Logit Mixing

Practical Tips

Practical tips and how-to guidance will be added by our editorial team.

中文解读

中文解读待补充：本站将优先为睡眠改善、失眠治疗、助眠方法等高价值文章补充中文说明。

Article Summary

Domain fine-tuning degrades the safety of large language models: fine-tuned specialists readily comply with harmful prompts framed in domain language. Existing inference-time defenses that mix logits from a safe anchor model require both models to share a vocabulary, which rules them out for the cross-family specialists where safety is most degraded. We present ALIGNBEAM, a training-free method that lifts this restriction by translating anchor logits into the target model's vocabulary token-by-token at each decoding step; a small LLM judge then selects the safest among K candidate continuations. No weights are changed, and the safety-utility trade-off can be tuned at deployment without retraining. Across both cross-vocabulary and same-vocabulary evaluation pairs, ALIGNBEAM substantially raises refusal on adversarial benchmarks while keeping task accuracy and inference overhead within practical bounds. The results show that safety alignment can be transferred between model families at inference time, without touching either model's weights.

Sources & References

• Original source
• Full article (PDF)

Comments