Mixture of Cluster-conditional LoRA Experts for Vision-language Instruction Tuning
This addresses performance issues in vision-language models for AI researchers, but it is incremental as it builds on existing MoE and LoRA methods.
The paper tackles task conflicts in instruction tuning of large vision-language models by proposing MoCLE, a mixture of experts architecture that activates task-customized parameters based on instruction clusters, and shows effectiveness in experiments on InstructBLIP and LLaVA.
Instruction tuning of Large Vision-language Models (LVLMs) has revolutionized the development of versatile models with zero-shot generalization across a wide range of downstream vision-language tasks. However, the diversity of training tasks of different sources and formats would lead to inevitable task conflicts, where different tasks conflict for the same set of model parameters, resulting in sub-optimal instruction-following abilities. To address that, we propose the Mixture of Cluster-conditional LoRA Experts (MoCLE), a novel Mixture of Experts (MoE) architecture designed to activate the task-customized model parameters based on the instruction clusters. A separate universal expert is further incorporated to improve generalization capabilities of MoCLE for novel instructions. Extensive experiments on InstructBLIP and LLaVA demonstrate the effectiveness of MoCLE.