Softmax Attention with Constant Cost per Token

Fuente: arXiv
Saved in:
Bibliographic Details
Main Author: Heinsen, Franz A.
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914773406318592
author Heinsen, Franz A.
author_facet Heinsen, Franz A.
contents We propose a simple modification to the conventional attention mechanism applied by Transformers: Instead of quantifying pairwise query-key similarity with scaled dot-products, we quantify it with the logarithms of scaled dot-products of exponentials. Our modification linearizes attention with exponential kernel feature maps, whose corresponding feature function is infinite dimensional. We show that our modification is expressible as a composition of log-sums of exponentials, with a latent space of constant size, enabling application with constant time and space complexity per token. We implement our modification, verify that it works in practice, and conclude that it is a promising alternative to conventional attention.
format Preprint
id arxiv_https___arxiv_org_abs_2404_05843
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Softmax Attention with Constant Cost per Token
Heinsen, Franz A.
Machine Learning
Computation and Language
We propose a simple modification to the conventional attention mechanism applied by Transformers: Instead of quantifying pairwise query-key similarity with scaled dot-products, we quantify it with the logarithms of scaled dot-products of exponentials. Our modification linearizes attention with exponential kernel feature maps, whose corresponding feature function is infinite dimensional. We show that our modification is expressible as a composition of log-sums of exponentials, with a latent space of constant size, enabling application with constant time and space complexity per token. We implement our modification, verify that it works in practice, and conclude that it is a promising alternative to conventional attention.
title Softmax Attention with Constant Cost per Token
topic Machine Learning
Computation and Language
url https://arxiv.org/abs/2404.05843