Enregistré dans:
Détails bibliographiques
Auteurs principaux: Shiratsuchi, Takumi, Tanaka, Yuichiro, Tamukoh, Hakaru
Format: Preprint
Publié: 2025
Sujets:
Accès en ligne:https://arxiv.org/abs/2512.23145
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866918265249333248
author Shiratsuchi, Takumi
Tanaka, Yuichiro
Tamukoh, Hakaru
author_facet Shiratsuchi, Takumi
Tanaka, Yuichiro
Tamukoh, Hakaru
contents Large language models (LLMs) have achieved state-of-the-art performance in natural language processing; however, their high computational cost remains a major bottleneck. In this study, we target computational efficiency by focusing on a matrix multiplication free language model (MatMul-free LM) and further reducing the training cost through an architecture inspired by reservoir computing. Specifically, we partially fix and share the weights of selected layers in the MatMul-free LM and insert reservoir layers to obtain rich dynamic representations without additional training overhead. Additionally, several operations are combined to reduce memory accesses. Experimental results show that the proposed architecture reduces the number of parameters by up to 19%, training time by 9.9%, and inference time by 8.0%, while maintaining comparable performance to the baseline model.
format Preprint
id arxiv_https___arxiv_org_abs_2512_23145
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Reservoir Computing inspired Matrix Multiplication-free Language Model
Shiratsuchi, Takumi
Tanaka, Yuichiro
Tamukoh, Hakaru
Computation and Language
Artificial Intelligence
Large language models (LLMs) have achieved state-of-the-art performance in natural language processing; however, their high computational cost remains a major bottleneck. In this study, we target computational efficiency by focusing on a matrix multiplication free language model (MatMul-free LM) and further reducing the training cost through an architecture inspired by reservoir computing. Specifically, we partially fix and share the weights of selected layers in the MatMul-free LM and insert reservoir layers to obtain rich dynamic representations without additional training overhead. Additionally, several operations are combined to reduce memory accesses. Experimental results show that the proposed architecture reduces the number of parameters by up to 19%, training time by 9.9%, and inference time by 8.0%, while maintaining comparable performance to the baseline model.
title Reservoir Computing inspired Matrix Multiplication-free Language Model
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2512.23145