SwitchCIT: Switching for Continual Instruction Tuning

TL;DR

SwitchCIT introduces a switching mechanism to mitigate catastrophic forgetting in continual instruction tuning of large models, enhancing efficiency, scalability, and task adaptability across language and vision-language tasks.

Contribution

The paper proposes a novel switching approach for continual instruction tuning that reduces forgetting and improves model adaptability with parameter-efficient tuning.

Findings

Effective in reducing catastrophic forgetting

Improves efficiency and scalability

Applicable to language and vision-language tasks

Abstract

Large language models (LLMs) and multimodal models (MMs) have exhibited impressive capabilities in various domains, particularly in general language understanding and visual reasoning. However, these models, trained on massive data, may not be finely optimized for specific tasks triggered by instructions. Continual instruction tuning is crucial to adapt a large model to evolving tasks and domains, ensuring their effectiveness and relevance across a wide range of applications. In the context of continual instruction tuning, where models are sequentially trained on different tasks, catastrophic forgetting can occur, leading to performance degradation on previously learned tasks. This work addresses the catastrophic forgetting in continual instruction learning through a switching mechanism for routing computations to parameter-efficient tuned models. We demonstrate the effectiveness of our method through experiments on continual instruction tuning of different natural language generation tasks and vision-language tasks. We also showcase the advantages of our proposed method in terms of efficiency, scalability, portability, and privacy preservation.