1+1>2: A Synergistic Sparse and Low-Rank Compression Method for Large Language Models
This provides a practical solution for efficient LLM deployment, addressing constraints in widespread adoption, though it is incremental as it builds on existing pruning and low-rank techniques.
The paper tackles the problem of compressing large language models (LLMs) to reduce bandwidth and computational demands by introducing a synergistic method combining sparse optimization and low-rank approximation, achieving a 50% compression on Qwen2.5 with no performance drop and at least 1.63x speedup.
Large Language Models (LLMs) have demonstrated remarkable proficiency in language comprehension and generation; however, their widespread adoption is constrained by substantial bandwidth and computational demands. While pruning and low-rank approximation have each demonstrated promising performance individually, their synergy for LLMs remains underexplored. We introduce \underline{S}ynergistic \underline{S}parse and \underline{L}ow-Rank \underline{C}ompression (SSLC) methods for LLMs, which leverages the strengths of both techniques: low-rank approximation compresses the model by retaining its essential structure with minimal information loss, whereas sparse optimization eliminates non-essential weights, preserving those crucial for generalization. Based on theoretical analysis, we first formulate the low-rank approximation and sparse optimization as a unified problem and solve it by iterative optimization algorithm. Experiments on LLaMA and Qwen2.5 models (7B-70B) show that SSLC, without any additional training steps, consistently surpasses standalone methods, achieving state-of-the-arts results. Notably, SSLC compresses Qwen2.5 by 50\% with no performance drop and achieves at least 1.63$\times$ speedup, offering a practical solution for efficient LLM deployment.