Saved in:
Bibliographic Details
Main Authors: Budzinskiy, Stanislav, Fang, Wenyi, Zeng, Longbin, Petersen, Philipp
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2503.10251
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912273523539968
author Budzinskiy, Stanislav
Fang, Wenyi
Zeng, Longbin
Petersen, Philipp
author_facet Budzinskiy, Stanislav
Fang, Wenyi
Zeng, Longbin
Petersen, Philipp
contents Large language models based on transformer architectures have become integral to state-of-the-art natural language processing applications. However, their training remains computationally expensive and exhibits instabilities, some of which are expected to be caused by finite-precision computations. We provide a theoretical analysis of the impact of round-off errors within the forward pass of a transformer architecture which yields fundamental bounds for these effects. In addition, we conduct a series of numerical experiments which demonstrate the practical relevance of our bounds. Our results yield concrete guidelines for choosing hyperparameters that mitigate round-off errors, leading to more robust and stable inference.
format Preprint
id arxiv_https___arxiv_org_abs_2503_10251
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Numerical Error Analysis of Large Language Models
Budzinskiy, Stanislav
Fang, Wenyi
Zeng, Longbin
Petersen, Philipp
Numerical Analysis
Machine Learning
Large language models based on transformer architectures have become integral to state-of-the-art natural language processing applications. However, their training remains computationally expensive and exhibits instabilities, some of which are expected to be caused by finite-precision computations. We provide a theoretical analysis of the impact of round-off errors within the forward pass of a transformer architecture which yields fundamental bounds for these effects. In addition, we conduct a series of numerical experiments which demonstrate the practical relevance of our bounds. Our results yield concrete guidelines for choosing hyperparameters that mitigate round-off errors, leading to more robust and stable inference.
title Numerical Error Analysis of Large Language Models
topic Numerical Analysis
Machine Learning
url https://arxiv.org/abs/2503.10251