SLoPe: Double-Pruned Sparse Plus Lazy Low-Rank Adapter Pretraining of LLMs

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Mozaffari, Mohammad, Yazdanbakhsh, Amir, Zhang, Zhao, Dehnavi, Maryam Mehri
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915121588076544
author Mozaffari, Mohammad
Yazdanbakhsh, Amir
Zhang, Zhao
Dehnavi, Maryam Mehri
author_facet Mozaffari, Mohammad
Yazdanbakhsh, Amir
Zhang, Zhao
Dehnavi, Maryam Mehri
contents We propose SLoPe, a Double-Pruned Sparse Plus Lazy Low-rank Adapter Pretraining method for LLMs that improves the accuracy of sparse LLMs while accelerating their pretraining and inference and reducing their memory footprint. Sparse pretraining of LLMs reduces the accuracy of the model, to overcome this, prior work uses dense models during fine-tuning. SLoPe improves the accuracy of sparsely pretrained models by adding low-rank adapters in the final 1% iterations of pretraining without adding significant overheads to the model pretraining and inference. In addition, SLoPe uses a double-pruned backward pass formulation that prunes the transposed weight matrix using N:M sparsity structures to enable an accelerated sparse backward pass. SLoPe accelerates the training and inference of models with billions of parameters up to $1.25\times$ and $1.54\times$ respectively (OPT-33B and OPT-66B) while reducing their memory usage by up to $0.63\times$ and $0.61\times$ for training and inference respectively.
format Preprint
id arxiv_https___arxiv_org_abs_2405_16325
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle SLoPe: Double-Pruned Sparse Plus Lazy Low-Rank Adapter Pretraining of LLMs
Mozaffari, Mohammad
Yazdanbakhsh, Amir
Zhang, Zhao
Dehnavi, Maryam Mehri
Machine Learning
Artificial Intelligence
We propose SLoPe, a Double-Pruned Sparse Plus Lazy Low-rank Adapter Pretraining method for LLMs that improves the accuracy of sparse LLMs while accelerating their pretraining and inference and reducing their memory footprint. Sparse pretraining of LLMs reduces the accuracy of the model, to overcome this, prior work uses dense models during fine-tuning. SLoPe improves the accuracy of sparsely pretrained models by adding low-rank adapters in the final 1% iterations of pretraining without adding significant overheads to the model pretraining and inference. In addition, SLoPe uses a double-pruned backward pass formulation that prunes the transposed weight matrix using N:M sparsity structures to enable an accelerated sparse backward pass. SLoPe accelerates the training and inference of models with billions of parameters up to $1.25\times$ and $1.54\times$ respectively (OPT-33B and OPT-66B) while reducing their memory usage by up to $0.63\times$ and $0.61\times$ for training and inference respectively.
title SLoPe: Double-Pruned Sparse Plus Lazy Low-Rank Adapter Pretraining of LLMs
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2405.16325