Query-based Cross-Modal Projector Bolstering Mamba Multimodal LLM
SooHwan Eom, Jay Shim, Gwanhyeong Koo, Haebin Na, Mark A. Hasegawa-Johnson, Sungwoong Kim, Chang D. Yoo
Abstract
The Transformer’s quadratic complexity with input length imposes an unsustainable computational load on large language models (LLMs). In contrast, the Selective Scan Structured State-Space Model, or Mamba, addresses this computational challenge effectively. This paper explores a query-based cross-modal projector designed to bolster Mamba’s efficiency for vision-language modeling by compressing visual tokens based on input through the cross-attention mechanism. This innovative projector also removes the need for manually designing the 2D scan order of original image features when converting them into an input sequence for Mamba LLM. Experimental results across various vision-language understanding benchmarks show that the proposed cross-modal projector enhances Mamba-based multimodal LLMs, boosting both performance and throughput.
BibTeX
@inproceedings{eom-etal-2024-query,
title = "Query-based Cross-Modal Projector Bolstering Mamba Multimodal {LLM}",
author = "Eom, SooHwan and
Shim, Jay and
Koo, Gwanhyeong and
Na, Haebin and
Hasegawa-Johnson, Mark A. and
Kim, Sungwoong and
Yoo, Chang D.",
editor = "Al-Onaizan, Yaser and
Bansal, Mohit and
Chen, Yun-Nung",
booktitle = "Findings of the Association for Computational Linguistics: EMNLP 2024",
month = nov,
year = "2024",
address = "Miami, Florida, USA",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2024.findings-emnlp.827/",
doi = "10.18653/v1/2024.findings-emnlp.827",
pages = "14158--14167"
}