MAGMA -- Multimodal Augmentation of Generative Models through Adapter-based Finetuning

TL;DR

MAGMA introduces a simple, adapter-based finetuning method that enhances generative language models with multimodal capabilities, achieving state-of-the-art results with minimal pretraining data and preserving language knowledge.

Contribution

It presents a novel end-to-end multimodal finetuning approach that maintains language model weights, enabling efficient training and transfer of pretraining knowledge.

Findings

Outperforms Frozen on open-ended generative tasks

Achieves state-of-the-art on OKVQA benchmark

Requires significantly less pretraining data

Abstract

Large-scale pretraining is fast becoming the norm in Vision-Language (VL) modeling. However, prevailing VL approaches are limited by the requirement for labeled data and the use of complex multi-step pretraining objectives. We present MAGMA - a simple method for augmenting generative language models with additional modalities using adapter-based finetuning. Building on Frozen, we train a series of VL models that autoregressively generate text from arbitrary combinations of visual and textual input. The pretraining is entirely end-to-end using a single language modeling objective, simplifying optimization compared to previous approaches. Importantly, the language model weights remain unchanged during training, allowing for transfer of encyclopedic knowledge and in-context learning abilities from language pretraining. MAGMA outperforms Frozen on open-ended generative tasks, achieving state of the art results on the OKVQA benchmark and competitive results on a range of other popular VL benchmarks, while pretraining on 0.2% of the number of samples used to train SimVLM.