OmniFusion Technical Report

TL;DR

OmniFusion is a multimodal AI model that combines large language models with visual adapters, achieving top performance on visual question answering benchmarks and supporting detailed domain-specific responses.

Contribution

The paper introduces OmniFusion, a novel multimodal architecture integrating pretrained LLMs with visual adapters, and demonstrates its superior performance across multiple visual-language benchmarks.

Findings

Achieved top scores on 8 visual-language benchmarks.

Demonstrated detailed responses in various domains like medicine and culture.

Open-source release of the OmniFusion model and training scripts.

Abstract

Last year, multimodal architectures served up a revolution in AI-based approaches and solutions, extending the capabilities of large language models (LLM). We propose an \textit{OmniFusion} model based on a pretrained LLM and adapters for visual modality. We evaluated and compared several architecture design principles for better text and visual data coupling: MLP and transformer adapters, various CLIP ViT-based encoders (SigLIP, InternVIT, etc.), and their fusing approach, image encoding method (whole image or tiles encoding) and two 7B LLMs (the proprietary one and open-source Mistral). Experiments on 8 visual-language benchmarks show the top score for the best OmniFusion setup in terms of different VQA tasks in comparison with open-source LLaVA-like solutions: VizWiz, Pope, MM-Vet, ScienceQA, MMBench, TextVQA, VQAv2, MMMU. We also propose a variety of situations, where OmniFusion provides highly-detailed answers in different domains: housekeeping, sightseeing, culture, medicine, handwritten and scanned equations recognition, etc. Mistral-based OmniFusion model is an open-source solution with weights, training and inference scripts available at https://github.com/AIRI-Institute/OmniFusion.