PIP: Perturbation-based Iterative Pruning for Large Language Models

TL;DR

PIP is a novel structured pruning method for large language models that reduces parameters by 20% while maintaining high accuracy, outperforming existing pruning techniques.

Contribution

Introduces PIP, a double-view structured pruning approach that leverages gradient differences to effectively prune LLMs with minimal performance loss.

Findings

Reduces model parameters by approximately 20%.

Maintains over 85% of original accuracy.

Outperforms existing state-of-the-art pruning methods.

Abstract

The rapid increase in the parameter counts of Large Language Models (LLMs), which often reach into the billions or even trillions, presents significant challenges for their practical deployment, particularly in resource-constrained environments. To address this issue, we propose PIP (Perturbation-based Iterative Pruning), a novel double-view structured pruning method to optimize LLMs, which combines information from two different views: the unperturbed view and the perturbed view. With the calculation of gradient differences, PIP iteratively prunes those that struggle to distinguish between these two views. Our experiments show that PIP reduces the parameter count by approximately 20% while retaining over 85% of the original model's accuracy across varied benchmarks. In some cases, the performance of the pruned model is within 5% of the unpruned version, demonstrating PIP's ability to preserve key aspects of model effectiveness. Moreover, PIP consistently outperforms existing state-of-the-art (SOTA) structured pruning methods, establishing it as a leading technique for optimizing LLMs in constrained environments.