IntraSlice: Towards High-Performance Structural Pruning with Block-Intra PCA for LLMs

TL;DR

IntraSlice introduces a block-wise PCA-based pruning method for LLMs that preserves model performance while significantly reducing size and inference time, by fully fusing transformation matrices without extra parameters.

Contribution

The paper proposes a novel intra-module PCA compression technique with a fully fusable transformation matrix and a global pruning estimator, improving structured pruning for LLMs.

Findings

Achieves higher compression with less performance loss.

Demonstrates superior results on Llama2, Llama3, and Phi models.

Outperforms recent baselines at equivalent compression ratios.

Abstract

Large Language Models (LLMs) achieve strong performance across diverse tasks but face deployment challenges due to their massive size. Structured pruning offers acceleration benefits but leads to significant performance degradation. Recent PCA-based pruning methods have alleviated this issue by retaining key activation components, but are only applied between modules in order to fuse the transformation matrix, which introduces extra parameters and severely disrupts activation distributions due to residual connections. To address these issues, we propose IntraSlice, a framework that applies block-wise module-intra PCA compression pruning. By leveraging the structural characteristics of Transformer modules, we design an approximate PCA method whose transformation matrices can be fully fused into the model without additional parameters. We also introduce a PCA-based global pruning ratio estimator that further considers the distribution of compressed activations, building on conventional module importance. We validate our method on Llama2, Llama3, and Phi series across various language benchmarks. Experimental results demonstrate that our approach achieves superior compression performance compared to recent baselines at the same compression ratio or inference speed.