Functional Interpolation for Relative Positions Improves Long Context Transformers

Preventing the performance decay of Transformers on inputs longer than those used for training has been an important challenge in extending the context length of these models. Though the Transformer architecture has fundamentally no limits on the input sequence lengths it can process, the choice of...

Full description

Saved in:
Bibliographic Details
Published in:arXiv.org 2024-03
Main Authors: Li, Shanda, You, Chong, Guru Guruganesh, Ainslie, Joshua, Ontanon, Santiago, Manzil Zaheer, Sanghai, Sumit, Yang, Yiming, Kumar, Sanjiv, Bhojanapalli, Srinadh
Format: Article
Language:English
Subjects:
Coding
Context
Interpolation
Training
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Preventing the performance decay of Transformers on inputs longer than those used for training has been an important challenge in extending the context length of these models. Though the Transformer architecture has fundamentally no limits on the input sequence lengths it can process, the choice of position encoding used during training can limit the performance of these models on longer inputs. We propose a novel functional relative position encoding with progressive interpolation, FIRE, to improve Transformer generalization to longer contexts. We theoretically prove that this can represent some of the popular relative position encodings, such as T5's RPE, Alibi, and Kerple. We next empirically show that FIRE models have better generalization to longer contexts on both zero-shot language modeling and long text benchmarks.
ISSN:2331-8422