Saved in:
Bibliographic Details
Main Authors: Wei, Xiuying, Moalla, Skander, Pascanu, Razvan, Gulcehre, Caglar
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2407.09835
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916334882783232
author Wei, Xiuying
Moalla, Skander
Pascanu, Razvan
Gulcehre, Caglar
author_facet Wei, Xiuying
Moalla, Skander
Pascanu, Razvan
Gulcehre, Caglar
contents State-of-the-art LLMs often rely on scale with high computational costs, which has sparked a research agenda to reduce parameter counts and costs without significantly impacting performance. Our study focuses on Transformer-based LLMs, specifically applying low-rank parametrization to the computationally intensive feedforward networks (FFNs), which are less studied than attention blocks. In contrast to previous works, (i) we explore low-rank parametrization at scale, up to 1.3B parameters; (ii) within Transformer language models rather than convolutional architectures; and (iii) starting from training from scratch. Experiments on the large RefinedWeb dataset show that low-rank parametrization is both efficient (e.g., 2.6$\times$ FFN speed-up with 32\% parameters) and effective during training. Interestingly, these structured FFNs exhibit steeper scaling curves than the original models. Motivated by this finding, we develop the wide and structured networks surpassing the current medium-sized and large-sized Transformer in perplexity and throughput performance. Our code is available at https://github.com/CLAIRE-Labo/StructuredFFN/tree/main.
format Preprint
id arxiv_https___arxiv_org_abs_2407_09835
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Investigating Low-Rank Training in Transformer Language Models: Efficiency and Scaling Analysis
Wei, Xiuying
Moalla, Skander
Pascanu, Razvan
Gulcehre, Caglar
Computation and Language
State-of-the-art LLMs often rely on scale with high computational costs, which has sparked a research agenda to reduce parameter counts and costs without significantly impacting performance. Our study focuses on Transformer-based LLMs, specifically applying low-rank parametrization to the computationally intensive feedforward networks (FFNs), which are less studied than attention blocks. In contrast to previous works, (i) we explore low-rank parametrization at scale, up to 1.3B parameters; (ii) within Transformer language models rather than convolutional architectures; and (iii) starting from training from scratch. Experiments on the large RefinedWeb dataset show that low-rank parametrization is both efficient (e.g., 2.6$\times$ FFN speed-up with 32\% parameters) and effective during training. Interestingly, these structured FFNs exhibit steeper scaling curves than the original models. Motivated by this finding, we develop the wide and structured networks surpassing the current medium-sized and large-sized Transformer in perplexity and throughput performance. Our code is available at https://github.com/CLAIRE-Labo/StructuredFFN/tree/main.
title Investigating Low-Rank Training in Transformer Language Models: Efficiency and Scaling Analysis
topic Computation and Language
url https://arxiv.org/abs/2407.09835