HyperLLaVA: Dynamic Visual and Language Expert Tuning for Multimodal Large Language Models

TL;DR

HyperLLaVA introduces a dynamic tuning approach for multimodal large language models, leveraging HyperNetworks to adapt visual and language experts, significantly improving performance on multiple benchmarks over static models.

Contribution

It proposes a novel dynamic tuning method using HyperNetworks for visual and language experts, surpassing static tuning strategies in multimodal large language models.

Findings

Outperforms LLaVA on multiple benchmarks

Demonstrates the effectiveness of adaptive expert tuning

Achieves significant performance improvements

Abstract

Recent advancements indicate that scaling up Multimodal Large Language Models (MLLMs) effectively enhances performance on downstream multimodal tasks. The prevailing MLLM paradigm, \emph{e.g.}, LLaVA, transforms visual features into text-like tokens using a \emph{static} vision-language mapper, thereby enabling \emph{static} LLMs to develop the capability to comprehend visual information through visual instruction tuning. Although promising, the \emph{static} tuning strategy~\footnote{The static tuning refers to the trained model with static parameters.} that shares the same parameters may constrain performance across different downstream multimodal tasks. In light of this, we introduce HyperLLaVA, which involves adaptive tuning of the projector and LLM parameters, in conjunction with a dynamic visual expert and language expert, respectively. These experts are derived from HyperNetworks, which generates adaptive parameter shifts through visual and language guidance, enabling dynamic projector and LLM modeling in two-stage training. Our experiments demonstrate that our solution significantly surpasses LLaVA on existing MLLM benchmarks, including MME, MMBench, SEED-Bench, and LLaVA-Bench. ~\footnote{Our project is available on the link https://github.com/DCDmllm/HyperLLaVA}.