Alignment-Constrained Dynamic Pruning for LLMs: Identifying and Preserving Alignment-Critical Circuits
This addresses the critical issue of maintaining safety in efficiently deployed LLMs, representing an incremental improvement over existing dynamic pruning methods.
The paper tackles the problem of alignment degradation in dynamically pruned large language models by introducing Alignment-Aware Probe Pruning (AAPP), which adaptively preserves alignment-relevant circuits during inference, resulting in a 50% improvement in refusal rates at matched computational cost across models like LLaMA 2-7B and Qwen2.5-14B-Instruct.
Large Language Models require substantial computational resources for inference, posing deployment challenges. While dynamic pruning offers superior efficiency over static methods through adaptive circuit selection, it exacerbates alignment degradation by retaining only input-dependent safety-critical circuit preservation across diverse inputs. As a result, addressing these heightened alignment vulnerabilities remains critical. We introduce Alignment-Aware Probe Pruning (AAPP), a dynamic structured pruning method that adaptively preserves alignment-relevant circuits during inference, building upon Probe Pruning. Experiments on LLaMA 2-7B, Qwen2.5-14B-Instruct, and Gemma-3-12B-IT show AAPP improves refusal rates by 50\% at matched compute, enabling efficient yet safety-preserving LLM deployment.