Improving Few-shot Generalization of Safety Classifiers via Data Augmented Parameter-Efficient Fine-Tuning
This addresses the challenge of few-shot generalization for safety classifiers in LLMs, which is crucial for adapting to emerging safety issues, though it is incremental as it builds on existing techniques like PEFT and data augmentation.
The paper tackles the problem of building safety classifiers for new, rarely observed safety rules in large language models by proposing a method that combines parameter-efficient fine-tuning with similarity-based data augmentation, achieving improvements of 7-17% F1 score in moral judgement and 9-13% AUC in toxicity detection tasks.
As large language models (LLMs) are widely adopted, new safety issues and policies emerge, to which existing safety classifiers do not generalize well. If we have only observed a few examples of violations of a new safety rule, how can we build a classifier to detect violations? In this paper, we study the novel setting of domain-generalized few-shot learning for LLM-based text safety classifiers. Unlike prior few-shot work, these new safety issues can be hard to uncover and we do not get to choose the few examples. We demonstrate that existing few-shot techniques do not perform well in this setting, and rather we propose to do parameter-efficient fine-tuning (PEFT) combined with augmenting training data based on similar examples in prior existing rules. We empirically show that our approach of similarity-based data-augmentation + prompt-tuning (DAPT) consistently outperforms baselines that either do not rely on data augmentation or on PEFT by 7-17% F1 score in the Social Chemistry moral judgement and 9-13% AUC in the Toxicity detection tasks, even when the new rule is loosely correlated with existing ones.