(Almost) Free Modality Stitching of Foundation Models

TL;DR

This paper introduces Hyma, a hypernetwork-based framework that significantly reduces the computational cost of selecting and aligning uni-modal models for multi-modal foundation models, achieving near grid search performance.

Contribution

Hyma provides an efficient, all-in-one hypernetwork approach for simultaneous uni-modal model selection and connector training, reducing search costs by tenfold.

Findings

Hyma reduces search cost by 10x.

Hyma matches grid search performance.

Effective across diverse benchmarks.

Abstract

Foundation multi-modal models are often designed by stitching of multiple existing pretrained uni-modal models: for example, an image classifier with an text model. This stitching process is performed by training a connector module that aims to align the representation spaces of these uni-modal models towards a multi-modal objective. However, given the complexity of training such connectors on large scale web-based datasets coupled with the ever-increasing number of available pretrained uni-modal models, the task of uni-modal models selection and subsequent connector module training becomes computationally demanding. To address this under-studied critical problem, we propose Hypernetwork Model Alignment (Hyma), a novel all-in-one solution for optimal uni-modal model selection and connector training by leveraging hypernetworks. Specifically, our framework utilizes the parameter prediction capability of a hypernetwork to obtain jointly trained connector modules for $N \times M$ combinations of uni-modal models. In our experiments, Hyma reduces the cost of searching for the best performing uni-modal model pair by $10\times$, while matching the ranking and trained connector performance obtained via grid search across a suite of diverse multi-modal benchmarks.