Maya: An Instruction Finetuned Multilingual Multimodal Model

TL;DR

Maya is an open-source multilingual multimodal model designed to improve understanding of low-resource languages and cultural nuances in vision-language tasks, addressing data quality and toxicity issues present in existing models.

Contribution

The paper introduces a multilingual image-text dataset, analyzes and mitigates toxicity in data, and develops a multilingual model supporting eight languages for better cultural and linguistic understanding.

Findings

Created a toxicity-free multilingual dataset

Enhanced model performance on low-resource languages

Demonstrated improved cultural understanding in vision-language tasks

Abstract

The rapid development of large Vision-Language Models (VLMs) has led to impressive results on academic benchmarks, primarily in widely spoken languages. However, significant gaps remain in the ability of current VLMs to handle low-resource languages and varied cultural contexts, largely due to a lack of high-quality, diverse, and safety-vetted data. Consequently, these models often struggle to understand low-resource languages and cultural nuances in a manner free from toxicity. To address these limitations, we introduce Maya, an open-source Multimodal Multilingual model. Our contributions are threefold: 1) a multilingual image-text pretraining dataset in eight languages, based on the LLaVA pretraining dataset; 2) a thorough analysis of toxicity within the LLaVA dataset, followed by the creation of a novel toxicity-free version across eight languages; and 3) a multilingual image-text model supporting these languages, enhancing cultural and linguistic comprehension in vision-language tasks. Code available at https://github.com/nahidalam/maya.