Modality Gap-Driven Subspace Alignment Training Paradigm For Multimodal Large Language Models

TL;DR

This paper introduces a geometric model of the modality gap in multimodal models and proposes ReAlign and ReVision strategies for efficient, scalable alignment using unpaired data, reducing reliance on costly image-text pairs.

Contribution

It offers a precise geometric characterization of the modality gap and introduces ReAlign and ReVision for effective, training-free alignment and scalable model training with unpaired data.

Findings

ReAlign effectively aligns text and image representations using unpaired data.

ReVision enables scalable training of multimodal models without large image-text datasets.

The proposed methods improve model alignment and scaling efficiency.

Abstract

Despite the success of multimodal contrastive learning in aligning visual and linguistic representations, a persistent geometric anomaly, the Modality Gap, remains: embeddings of distinct modalities expressing identical semantics occupy systematically offset regions. Prior approaches to bridge this gap are largely limited by oversimplified isotropic assumptions, hindering their application in large-scale scenarios. In this paper, we address these limitations by precisely characterizing the geometric shape of the modality gap and leveraging it for efficient model scaling. First, we propose the Fixed-frame Modality Gap Theory, which decomposes the modality gap within a frozen reference frame into stable biases and anisotropic residuals. Guided by this precise modeling, we introduce ReAlign, a training-free modality alignment strategy. Utilizing statistics from massive unpaired data, ReAlign aligns text representation into the image representation distribution via a three-step process comprising Anchor, Trace, and Centroid Alignment, thereby explicitly rectifying geometric misalignment. Building on ReAlign, we propose ReVision, a scalable training paradigm for Multimodal Large Language Models~(MLLMs). ReVision integrates ReAlign into the pretraining stage, enabling the model to learn the distribution of visual representations from unpaired text before visual instruction tuning, without the need for large-scale, high-quality image-text pairs. Our framework demonstrates that statistically aligned unpaired data can effectively substitute for expensive image-text pairs, offering a robust path for the efficient scaling of MLLMs.