Steering When Necessary: Flexible Steering Large Language Models with Backtracking

TL;DR

The paper introduces FASB, a dynamic activation steering framework with backtracking for LLMs, improving alignment with desired behaviors during inference without costly fine-tuning.

Contribution

It proposes a novel method that adaptively determines when and how strongly to intervene in LLMs during generation, incorporating backtracking to correct deviations.

Findings

Outperforms baseline methods on TruthfulQA and six multiple-choice datasets.

Effectively balances intervention strength with minimal disruption.

Demonstrates improved alignment with desired behaviors during inference.

Abstract

Large language models (LLMs) have achieved remarkable performance across many generation tasks. Nevertheless, effectively aligning them with desired behaviors remains a significant challenge. Activation steering is an effective and cost-efficient approach that directly modifies the activations of LLMs during the inference stage, aligning their responses with the desired behaviors and avoiding the high cost of fine-tuning. Existing methods typically indiscriminately intervene to all generations or rely solely on the question to determine intervention, which limits the accurate assessment of the intervention strength. To this end, we propose the Flexible Activation Steering with Backtracking (FASB) framework, which dynamically determines both the necessity and strength of intervention by tracking the internal states of the LLMs during generation, considering both the question and the generated content. Since intervening after detecting a deviation from the desired behavior is often too late, we further propose the backtracking mechanism to correct the deviated tokens and steer the LLMs toward the desired behavior. Extensive experiments on the TruthfulQA dataset and six multiple-choice datasets demonstrate that our method outperforms baselines. Our code will be released at https://github.com/gjw185/FASB.