MobileVLM V2: Faster and Stronger Baseline for Vision Language Model

Xiangxiang Chu, Limeng Qiao, Xinyu Zhang, Shuang Xu, Fei Wei +6 more
2/6/2024
cs.CVcs.AI

Official code: Meituan-AutoML/MobileVLM (1,365 stars, Apache-2.0, last commit 2024-04-15).

Abstract

We introduce MobileVLM V2, a family of significantly improved vision language models upon MobileVLM, which proves that a delicate orchestration of novel architectural design, an improved training scheme tailored for mobile VLMs, and rich high-quality dataset curation can substantially benefit VLMs' performance. Specifically, MobileVLM V2 1.7B achieves better or on-par performance on standard VLM benchmarks compared with much larger VLMs at the 3B scale. Notably, our 3B model outperforms a large variety of VLMs at the 7B+ scale. Our models will be released at https://github.com/Meituan-AutoML/MobileVLM .

View on arXivView PDF

Code Implementations(1)

Meituan-AutoML/MobileVLMOfficial100%
Apache-2.0

Strong and Open Vision Language Assistant for Mobile Devices

1,36589Shell, PythonDec 29, 20232 years agoApache-2.0

Similar Papers

Eve: Efficient Multimodal Vision Language Models with Elastic Visual Experts

90% match1 repo132025

A Survey on Efficient Vision-Language Models

89% match2 repos3.1k2025

POINTS: Improving Your Vision-language Model with Affordable Strategies

89% match2024

Building and better understanding vision-language models: insights and future directions

89% match2024

What matters when building vision-language models?

88% match2026

Meteor: Mamba-based Traversal of Rationale for Large Language and Vision Models

88% match1 repo1172024

Cite this paper

@article{chu2024mobilevlm,
  title  = {MobileVLM V2: Faster and Stronger Baseline for Vision Language Model},
  author = {Xiangxiang Chu and Limeng Qiao and Xinyu Zhang and Shuang Xu and Fei Wei and Yang Yang and Xiaofei Sun and Yiming Hu and Xinyang Lin and Bo Zhang and Chunhua Shen},
  year   = {2024},
  eprint = {2402.03766},
  archivePrefix = {arXiv},
  url    = {http://arxiv.org/abs/2402.03766v1}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/2402.03766.svg)](https://codeofpaper.com/paper/2402.03766)

Discussion