Multi-Modal Generative Embedding Model

TL;DR

This paper introduces MM-GEM, a unified multi-modal model that combines generation and embedding tasks within a single large language model, achieving competitive results across various benchmarks.

Contribution

The work presents a minimalistic multi-modal paradigm with a single model for both generation and embedding, using a novel PoolAggregator for efficiency and fine-grained tasks.

Findings

Competitive performance on cross-modal retrieval and zero-shot classification

Effective region-level image captioning and retrieval

Over 5% improvement in Recall@1 for long text and image retrieval

Abstract

Most multi-modal tasks can be formulated into problems of either generation or embedding. Existing models usually tackle these two types of problems by decoupling language modules into a text decoder for generation, and a text encoder for embedding. To explore the minimalism of multi-modal paradigms, we attempt to achieve only one model per modality in this work. We propose a Multi-Modal Generative Embedding Model (MM-GEM), whereby the generative and embedding objectives are encapsulated in one Large Language Model. We also propose a PoolAggregator to boost efficiency and enable the ability of fine-grained embedding and generation. A surprising finding is that these two objectives do not significantly conflict with each other. For example, MM-GEM instantiated from ViT-Large and TinyLlama shows competitive performance on benchmarks for multimodal embedding models such as cross-modal retrieval and zero-shot classification, while has good ability of image captioning. Additionally, MM-GEM can seamlessly execute region-level image caption generation and retrieval tasks. Besides, the advanced text model in MM-GEM brings over 5% improvement in Recall@1 for long text and image retrieval.