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.
@article{arxiv.2404.08567,
title = {CATP: Cross-Attention Token Pruning for Accuracy Preserved Multimodal Model Inference},
author = {Ruqi Liao and Chuqing Zhao and Jin Li and Weiqi Feng and Yi Lyu and Bingxian Chen and Haochen Yang},
journal= {arXiv preprint arXiv:2404.08567},
year = {2026}
}