Saved in:
Bibliographic Details
Main Authors: Liao, Ruqi, Zhao, Chuqing, Li, Jin, Feng, Weiqi, Lyu, Yi, Chen, Bingxian, Yang, Haochen
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2404.08567
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910020638081024
author Liao, Ruqi
Zhao, Chuqing
Li, Jin
Feng, Weiqi
Lyu, Yi
Chen, Bingxian
Yang, Haochen
author_facet Liao, Ruqi
Zhao, Chuqing
Li, Jin
Feng, Weiqi
Lyu, Yi
Chen, Bingxian
Yang, Haochen
contents In response to the rising interest in large multimodal models, we introduce Cross-Attention Token Pruning (CATP), a precision-focused token pruning method. Our approach leverages cross-attention layers in multimodal models, exemplified by BLIP-2, to extract valuable information for token importance determination. CATP employs a refined voting strategy across model heads and layers. In evaluations, CATP achieves up to 12.1X higher accuracy compared to existing token pruning methods, addressing the trade-off between computational efficiency and model precision.
format Preprint
id arxiv_https___arxiv_org_abs_2404_08567
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle CATP: Cross-Attention Token Pruning for Accuracy Preserved Multimodal Model Inference
Liao, Ruqi
Zhao, Chuqing
Li, Jin
Feng, Weiqi
Lyu, Yi
Chen, Bingxian
Yang, Haochen
Computation and Language
Artificial Intelligence
In response to the rising interest in large multimodal models, we introduce Cross-Attention Token Pruning (CATP), a precision-focused token pruning method. Our approach leverages cross-attention layers in multimodal models, exemplified by BLIP-2, to extract valuable information for token importance determination. CATP employs a refined voting strategy across model heads and layers. In evaluations, CATP achieves up to 12.1X higher accuracy compared to existing token pruning methods, addressing the trade-off between computational efficiency and model precision.
title CATP: Cross-Attention Token Pruning for Accuracy Preserved Multimodal Model Inference
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2404.08567