mBLIP: Efficient Bootstrapping of Multilingual Vision-LLMs

TL;DR

mBLIP introduces a computationally efficient method to adapt multilingual vision-language models using machine-translated data, achieving competitive results without extensive end-to-end multilingual pretraining.

Contribution

The paper presents the first multilingual Vision-LLM that is efficiently aligned with multilingual LLMs using limited data and without costly pretraining.

Findings

mBLIP performs competitively on IGLUE and XM3600 benchmarks.

It significantly outperforms English-only Vision-LLMs like Llava 1.5.

The approach reduces computational costs for multilingual Vision-LLMs.

Abstract

Modular vision-language models (Vision-LLMs) align pretrained image encoders with (frozen) large language models (LLMs) and post-hoc condition LLMs to `understand' the image input. With the abundance of readily available high-quality English image-text data as well as strong monolingual English LLMs, the research focus has been on English-only Vision-LLMs. Multilingual vision-language models are still predominantly obtained via expensive end-to-end pretraining, resulting in comparatively smaller models, trained on limited multilingual image data supplemented with text-only multilingual corpora. We present mBLIP, the first Vision-LLM leveraging multilingual LLMs, which we obtain in a computationally efficient manner on consumer-level hardware. To this end, we \textit{re-align} an image encoder previously tuned to an English LLM to a new, multilingual LLM using only a few million multilingual training examples derived from a mix of vision-and-language tasks, which we obtain by machine-translating high-quality English data to 95 languages. On the IGLUE benchmark and XM3600, mBLIP yields results competitive with state-of-the-art models and it greatly outperforms strong English-only Vision-LLMs like Llava 1.5. We release our model, code, and train data at \url{https://github.com/gregor-ge/mBLIP}.