@inproceedings{zhu-etal-2024-coca,
title = "{C}o{CA}: Fusing Position Embedding with Collinear Constrained Attention in Transformers for Long Context Window Extending",
author = "Zhu, Shiyi and
Ye, Jing and
Jiang, Wei and
Xue, Siqiao and
Zhang, Qi and
Wu, Yifan and
Li, Jianguo",
editor = "Ku, Lun-Wei and
Martins, Andre and
Srikumar, Vivek",
booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)",
month = aug,
year = "2024",
address = "Bangkok, Thailand",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2024.luhme-long.233/",
doi = "10.18653/v1/2024.acl-long.233",
pages = "4247--4262",
abstract = "Self-attention and position embedding are two crucial modules in transformer-based Large Language Models (LLMs). However, the potential relationship between them is far from well studied, especially for long context window extending. In fact, anomalous behaviors that hinder long context extrapolation exist between Rotary Position Embedding (RoPE) and vanilla self-attention.Incorrect initial angles between $Q$ and $K$ can cause misestimation in modeling rotary position embedding of the closest tokens.To address this issue, we propose $\textbf{Co}$llinear $\textbf{C}$onstrained $\textbf{A}$ttention mechanism, namely CoCA. Specifically, we enforce a collinear constraint between $Q$ and $K$ to seamlessly integrate RoPE and self-attention.While only adding minimal computational and spatial complexity, this integration significantly enhances long context window extrapolation ability. We provide an optimized implementation, making it a drop-in replacement for any existing transformer-based models.Extensive experiments demonstrate that CoCA excels in extending context windows. A CoCA-based GPT model, trained with a context length of 512, can extend the context window up to 32K (60$\times$) without any fine-tuning.Additionally, incorporating CoCA into LLaMA-7B achieves extrapolation up to 32K within a training length of only 2K.Our code is publicly available at: https://github.com/codefuse-ai/Collinear-Constrained-Attention"
}
<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="zhu-etal-2024-coca">
<titleInfo>
<title>CoCA: Fusing Position Embedding with Collinear Constrained Attention in Transformers for Long Context Window Extending</title>
</titleInfo>
<name type="personal">
<namePart type="given">Shiyi</namePart>
<namePart type="family">Zhu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jing</namePart>
<namePart type="family">Ye</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Wei</namePart>
<namePart type="family">Jiang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Siqiao</namePart>
<namePart type="family">Xue</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Qi</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yifan</namePart>
<namePart type="family">Wu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jianguo</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2024-08</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Lun-Wei</namePart>
<namePart type="family">Ku</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Andre</namePart>
<namePart type="family">Martins</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Vivek</namePart>
<namePart type="family">Srikumar</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Bangkok, Thailand</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
</relatedItem>
<abstract>Self-attention and position embedding are two crucial modules in transformer-based Large Language Models (LLMs). However, the potential relationship between them is far from well studied, especially for long context window extending. In fact, anomalous behaviors that hinder long context extrapolation exist between Rotary Position Embedding (RoPE) and vanilla self-attention.Incorrect initial angles between Q and K can cause misestimation in modeling rotary position embedding of the closest tokens.To address this issue, we propose Collinear Constrained Attention mechanism, namely CoCA. Specifically, we enforce a collinear constraint between Q and K to seamlessly integrate RoPE and self-attention.While only adding minimal computational and spatial complexity, this integration significantly enhances long context window extrapolation ability. We provide an optimized implementation, making it a drop-in replacement for any existing transformer-based models.Extensive experiments demonstrate that CoCA excels in extending context windows. A CoCA-based GPT model, trained with a context length of 512, can extend the context window up to 32K (60\times) without any fine-tuning.Additionally, incorporating CoCA into LLaMA-7B achieves extrapolation up to 32K within a training length of only 2K.Our code is publicly available at: https://github.com/codefuse-ai/Collinear-Constrained-Attention</abstract>
<identifier type="citekey">zhu-etal-2024-coca</identifier>
<identifier type="doi">10.18653/v1/2024.acl-long.233</identifier>
<location>
<url>https://aclanthology.org/2024.luhme-long.233/</url>
</location>
<part>
<date>2024-08</date>
<extent unit="page">
<start>4247</start>
<end>4262</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T CoCA: Fusing Position Embedding with Collinear Constrained Attention in Transformers for Long Context Window Extending
%A Zhu, Shiyi
%A Ye, Jing
%A Jiang, Wei
%A Xue, Siqiao
%A Zhang, Qi
%A Wu, Yifan
%A Li, Jianguo
%Y Ku, Lun-Wei
%Y Martins, Andre
%Y Srikumar, Vivek
%S Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)
%D 2024
%8 August
%I Association for Computational Linguistics
%C Bangkok, Thailand
%F zhu-etal-2024-coca
%X Self-attention and position embedding are two crucial modules in transformer-based Large Language Models (LLMs). However, the potential relationship between them is far from well studied, especially for long context window extending. In fact, anomalous behaviors that hinder long context extrapolation exist between Rotary Position Embedding (RoPE) and vanilla self-attention.Incorrect initial angles between Q and K can cause misestimation in modeling rotary position embedding of the closest tokens.To address this issue, we propose Collinear Constrained Attention mechanism, namely CoCA. Specifically, we enforce a collinear constraint between Q and K to seamlessly integrate RoPE and self-attention.While only adding minimal computational and spatial complexity, this integration significantly enhances long context window extrapolation ability. We provide an optimized implementation, making it a drop-in replacement for any existing transformer-based models.Extensive experiments demonstrate that CoCA excels in extending context windows. A CoCA-based GPT model, trained with a context length of 512, can extend the context window up to 32K (60\times) without any fine-tuning.Additionally, incorporating CoCA into LLaMA-7B achieves extrapolation up to 32K within a training length of only 2K.Our code is publicly available at: https://github.com/codefuse-ai/Collinear-Constrained-Attention
%R 10.18653/v1/2024.acl-long.233
%U https://aclanthology.org/2024.luhme-long.233/
%U https://doi.org/10.18653/v1/2024.acl-long.233
%P 4247-4262
Markdown (Informal)
[CoCA: Fusing Position Embedding with Collinear Constrained Attention in Transformers for Long Context Window Extending](https://aclanthology.org/2024.luhme-long.233/) (Zhu et al., ACL 2024)
ACL