Data-driven Clustering and Merging of Adapters for On-device Large Language Models

TL;DR

This paper introduces D2C, a novel clustering method for adapters in large language models that uses minimal data and iterative optimization to create multi-task adapters suitable for on-device deployment, improving performance within storage limits.

Contribution

The paper presents a new adapter clustering technique that efficiently merges task-specific adapters using minimal examples and iterative refinement, enabling effective multi-task adapters for resource-limited devices.

Findings

Effective adapter clustering with minimal data

Improved multi-task adapter performance

Suitable for deployment on resource-constrained devices

Abstract

On-device large language models commonly employ task-specific adapters (e.g., LoRAs) to deliver strong performance on downstream tasks. While storing all available adapters is impractical due to memory constraints, mobile devices typically have sufficient capacity to store a limited number of these parameters. This raises a critical challenge: how to select representative adapters that generalize well across multiple tasks - a problem that remains unexplored in existing literature. We propose a novel method D2C for adapter clustering that leverages minimal task-specific examples (e.g., 10 per task) and employs an iterative optimization process to refine cluster assignments. The adapters within each cluster are merged, creating multi-task adapters deployable on resource-constrained devices. Experimental results demonstrate that our method effectively boosts performance for considered storage budgets.