EAGLE: EXPLORING THE DESIGN SPACE FOR MULTI- MODAL LLMS WITH MIXTURE OF ENCODERS

• Article Name: EAGLE: EXPLORING THE DESIGN SPACE FOR MULTI-

MODAL LLMS WITH MIXTURE OF ENCODERS

• Topic: Development and training of multimodal AI models (Vision-Language Models - VLMs) for long-context video and high-resolution image understanding, using a technique called "expertise-aware mixture-of-experts" to improve performance.

• Year: 2024

• Link: https://arxiv.org/pdf/2408.15998

• Students: Yarden Barel, Maroun Nasra

https://github.com/NVlabs/EAGLE/tree/main/Eagle1