From Pruning to Grafting: Dynamic Knowledge Redistribution via Learnable Layer Fusion

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Pei, Zehua, Zhen, Hui-Ling, Yu, Xianzhi, Pan, Sinno Jialin, Yuan, Mingxuan, Yu, Bei
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909960911192064
author Pei, Zehua
Zhen, Hui-Ling
Yu, Xianzhi
Pan, Sinno Jialin
Yuan, Mingxuan
Yu, Bei
author_facet Pei, Zehua
Zhen, Hui-Ling
Yu, Xianzhi
Pan, Sinno Jialin
Yuan, Mingxuan
Yu, Bei
contents Structured pruning of Generative Pre-trained Transformers (GPTs) offers a promising path to efficiency but often suffers from irreversible performance degradation due to the discarding of transformer blocks. In this paper, we introduce FuseGPT, a compression paradigm that reframes structured pruning as iterative knowledge grafting rather than simple removal. Motivated by the observation that linear block merging fails to capture non-linear feature disparities and that block importance fluctuates dynamically during pruning, FuseGPT employs a dual-strategy pipeline. First, we propose Macro Influence (MI), a dynamic fusion-aware metric that continuously re-evaluates block redundancy as the network topology evolves. Second, instead of rigid parameter averaging, we introduce a learnable low-rank fusion mechanism that adaptively grafts the knowledge of pruned blocks onto surviving layers via lightweight local distillation. Extensive experiments on LLaMA, Mistral, Qwen, and Phi families demonstrate that FuseGPT establishes a new state-of-the-art on the compression-accuracy Pareto frontier: at 25\% sparsity, FuseGPT achieves lower perplexity than prior methods at 20\% sparsity, improves zero-shot reasoning by up to 4.5 points, and delivers 1.33$\times$ inference speedup with 25\% memory reduction. Furthermore, FuseGPT is orthogonal to quantization, achieving 52.1\% total compression with negligible quality loss when combined with 4-bit GPTQ. We make our code publicly available at https://github.com/JarvisPei/FuseGPT.
format Preprint
id arxiv_https___arxiv_org_abs_2411_14507
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle From Pruning to Grafting: Dynamic Knowledge Redistribution via Learnable Layer Fusion
Pei, Zehua
Zhen, Hui-Ling
Yu, Xianzhi
Pan, Sinno Jialin
Yuan, Mingxuan
Yu, Bei
Machine Learning
Artificial Intelligence
Computation and Language
Structured pruning of Generative Pre-trained Transformers (GPTs) offers a promising path to efficiency but often suffers from irreversible performance degradation due to the discarding of transformer blocks. In this paper, we introduce FuseGPT, a compression paradigm that reframes structured pruning as iterative knowledge grafting rather than simple removal. Motivated by the observation that linear block merging fails to capture non-linear feature disparities and that block importance fluctuates dynamically during pruning, FuseGPT employs a dual-strategy pipeline. First, we propose Macro Influence (MI), a dynamic fusion-aware metric that continuously re-evaluates block redundancy as the network topology evolves. Second, instead of rigid parameter averaging, we introduce a learnable low-rank fusion mechanism that adaptively grafts the knowledge of pruned blocks onto surviving layers via lightweight local distillation. Extensive experiments on LLaMA, Mistral, Qwen, and Phi families demonstrate that FuseGPT establishes a new state-of-the-art on the compression-accuracy Pareto frontier: at 25\% sparsity, FuseGPT achieves lower perplexity than prior methods at 20\% sparsity, improves zero-shot reasoning by up to 4.5 points, and delivers 1.33$\times$ inference speedup with 25\% memory reduction. Furthermore, FuseGPT is orthogonal to quantization, achieving 52.1\% total compression with negligible quality loss when combined with 4-bit GPTQ. We make our code publicly available at https://github.com/JarvisPei/FuseGPT.
title From Pruning to Grafting: Dynamic Knowledge Redistribution via Learnable Layer Fusion
topic Machine Learning
Artificial Intelligence
Computation and Language
url https://arxiv.org/abs/2411.14507