MeteoRA: Multiple-tasks Embedded LoRA for Large Language Models

TL;DR

MeteoRA introduces a scalable Mixture-of-Experts framework to embed multiple task-specific LoRA adapters into large language models, enabling efficient multi-task handling and autonomous task switching during inference.

Contribution

The paper proposes MeteoRA, a novel framework that embeds multiple LoRA adapters into LLMs using MoE architecture, improving multi-task performance and adapter switching efficiency.

Findings

Achieves equivalent performance to traditional PEFT methods.

Enables LLMs to handle multiple tasks in a single inference pass.

Demonstrates superior performance in composite task scenarios.

Abstract

The pretrain+fine-tune paradigm is foundational for deploying large language models (LLMs) across various downstream applications. Within this framework, Low-Rank Adaptation (LoRA) stands out for its parameter-efficient fine-tuning (PEFT), producing numerous reusable task-specific LoRA adapters. However, this approach requires explicit task intention selection, posing challenges for autonomous task sensing and switching during inference with multiple existing LoRA adapters embedded in a single LLM. In this work, we introduce MeteoRA (Multiple-tasks embedded LoRA), a scalable and efficient framework that reuses multiple task-specific LoRA adapters into the base LLM via a full-mode Mixture-of-Experts (MoE) architecture. This framework also includes novel MoE forward acceleration strategies to address the efficiency challenges of traditional MoE implementations. Our evaluation, using the LlaMA2-13B and LlaMA3-8B base models equipped with 28 existing LoRA adapters through MeteoRA, demonstrates equivalent performance with the traditional PEFT method. Moreover, the LLM equipped with MeteoRA achieves superior performance in handling composite tasks, effectively solving ten sequential problems in a single inference pass, thereby demonstrating the framework's enhanced capability for timely adapter switching.