Aria: An Open Multimodal Native Mixture-of-Experts Model

TL;DR

Aria is an open, multimodal native mixture-of-experts model that achieves state-of-the-art performance across diverse tasks, promoting accessibility and adaptability in multimodal AI research and applications.

Contribution

Introduces Aria, a large open-source multimodal mixture-of-experts model with competitive performance, filling the gap of proprietary models and enabling broader adoption.

Findings

Outperforms Pixtral-12B and Llama3.2-11B on various tasks

Achieves best-in-class performance across multimodal, language, and coding tasks

Open-sourced model weights and codebase for easy adoption

Abstract

Information comes in diverse modalities. Multimodal native AI models are essential to integrate real-world information and deliver comprehensive understanding. While proprietary multimodal native models exist, their lack of openness imposes obstacles for adoptions, let alone adaptations. To fill this gap, we introduce Aria, an open multimodal native model with best-in-class performance across a wide range of multimodal, language, and coding tasks. Aria is a mixture-of-expert model with 3.9B and 3.5B activated parameters per visual token and text token, respectively. It outperforms Pixtral-12B and Llama3.2-11B, and is competitive against the best proprietary models on various multimodal tasks. We pre-train Aria from scratch following a 4-stage pipeline, which progressively equips the model with strong capabilities in language understanding, multimodal understanding, long context window, and instruction following. We open-source the model weights along with a codebase that facilitates easy adoptions and adaptations of Aria in real-world applications.