GRASP: Replace Redundant Layers with Adaptive Singular Parameters for Efficient Model Compression

TL;DR

GRASP introduces a gradient-based method that adaptively retains critical singular parameters to compress large language models efficiently, reducing redundancy while preserving performance.

Contribution

It proposes a novel gradient-based framework that replaces redundant layers with minimal adaptive singular parameters, improving compression without significant performance loss.

Findings

Achieves 90% of original performance at 20% compression ratio.

Outperforms existing compression methods across multiple LLMs.

Uses gradient attribution on small calibration datasets for adaptive layer replacement.

Abstract

Recent studies have demonstrated that many layers are functionally redundant in large language models (LLMs), enabling model compression by removing these layers to reduce inference cost. While such approaches can improve efficiency, indiscriminate layer pruning often results in significant performance degradation. In this paper, we propose GRASP (Gradient-based Retention of Adaptive Singular Parameters), a novel compression framework that mitigates this issue by preserving sensitivity-aware singular values. Unlike direct layer pruning, GRASP leverages gradient-based attribution on a small calibration dataset to adaptively identify and retain critical singular components. By replacing redundant layers with only a minimal set of parameters, GRASP achieves efficient compression while maintaining strong performance with minimal overhead. Experiments across multiple LLMs show that GRASP consistently outperforms existing compression methods, achieving 90% of the original model's performance under 20% compression ratio.