Saved in:
Bibliographic Details
Main Authors: Chen, Yinpeng, Hutchins, DeLesley, Jansen, Aren, Zhmoginov, Andrey, Racz, David, Andersen, Jesper
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2410.03156
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909335824629760
author Chen, Yinpeng
Hutchins, DeLesley
Jansen, Aren
Zhmoginov, Andrey
Racz, David
Andersen, Jesper
author_facet Chen, Yinpeng
Hutchins, DeLesley
Jansen, Aren
Zhmoginov, Andrey
Racz, David
Andersen, Jesper
contents We present MELODI, a novel memory architecture designed to efficiently process long documents using short context windows. The key principle behind MELODI is to represent short-term and long-term memory as a hierarchical compression scheme across both network layers and context windows. Specifically, the short-term memory is achieved through recurrent compression of context windows across multiple layers, ensuring smooth transitions between windows. In contrast, the long-term memory performs further compression within a single middle layer and aggregates information across context windows, effectively consolidating crucial information from the entire history. Compared to a strong baseline - the Memorizing Transformer employing dense attention over a large long-term memory (64K key-value pairs) - our method demonstrates superior performance on various long-context datasets while remarkably reducing the memory footprint by a factor of 8.
format Preprint
id arxiv_https___arxiv_org_abs_2410_03156
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle MELODI: Exploring Memory Compression for Long Contexts
Chen, Yinpeng
Hutchins, DeLesley
Jansen, Aren
Zhmoginov, Andrey
Racz, David
Andersen, Jesper
Machine Learning
Artificial Intelligence
We present MELODI, a novel memory architecture designed to efficiently process long documents using short context windows. The key principle behind MELODI is to represent short-term and long-term memory as a hierarchical compression scheme across both network layers and context windows. Specifically, the short-term memory is achieved through recurrent compression of context windows across multiple layers, ensuring smooth transitions between windows. In contrast, the long-term memory performs further compression within a single middle layer and aggregates information across context windows, effectively consolidating crucial information from the entire history. Compared to a strong baseline - the Memorizing Transformer employing dense attention over a large long-term memory (64K key-value pairs) - our method demonstrates superior performance on various long-context datasets while remarkably reducing the memory footprint by a factor of 8.
title MELODI: Exploring Memory Compression for Long Contexts
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2410.03156