AIM: Let Any Multi-modal Large Language Models Embrace Efficient In-Context Learning

TL;DR

AIM introduces a lightweight framework that enables multi-modal large language models to perform efficient in-context learning by converting visual information into fused virtual tokens, reducing reliance on multi-modal training data.

Contribution

The paper proposes a novel method to incorporate multi-modal demonstrations into LLMs by transforming image-text pairs into fused tokens, enhancing multi-modal ICL without retraining the core model.

Findings

AIM effectively enables multi-modal ICL with minimal additional training.

The framework improves performance on multi-modal tasks without modifying the LLM.

AIM is compatible with any frozen MLLM and trained on public multi-modal data.

Abstract

In-context learning (ICL) facilitates Large Language Models (LLMs) exhibiting emergent ability on downstream tasks without updating billions of parameters. However, in the area of multi-modal Large Language Models (MLLMs), two problems hinder the application of multi-modal ICL: (1) Most primary MLLMs are only trained on single-image datasets, making them unable to read multi-modal demonstrations. (2) With the demonstrations increasing, thousands of visual tokens highly challenge hardware and degrade ICL performance. During preliminary explorations, we discovered that the inner LLM tends to focus more on the linguistic modality within multi-modal demonstrations to generate responses. Therefore, we propose a general and light-weighted framework \textbf{AIM} to tackle the mentioned problems through \textbf{A}ggregating \textbf{I}mage information of \textbf{M}ultimodal demonstrations to the dense latent space of the corresponding linguistic part. Specifically, AIM first uses the frozen backbone MLLM to read each image-text demonstration and extracts the vector representations on top of the text. These vectors naturally fuse the information of the image-text pair, and AIM transforms them into fused virtual tokens acceptable for the inner LLM via a trainable projection layer. Ultimately, these fused tokens function as variants of multi-modal demonstrations, fed into the MLLM to direct its response to the current query as usual. Because these fused tokens stem from the textual component of the image-text pair, a multi-modal demonstration is nearly reduced to a pure textual demonstration, thus seamlessly applying to any MLLMs. With its de facto MLLM frozen, AIM is parameter-efficient and we train it on public multi-modal web corpora which have nothing to do with downstream test tasks.