MambaPEFT: Exploring Parameter-Efficient Fine-Tuning for Mamba

TL;DR

This paper explores parameter-efficient fine-tuning (PEFT) methods for Mamba, a state space model alternative to Transformers, demonstrating improved adaptation to downstream tasks and proposing Mamba-specific PEFT techniques.

Contribution

It introduces Mamba-specific PEFT methods, modifies existing PEFT techniques for Mamba, and provides a framework that outperforms previous approaches in adapting Mamba models.

Findings

PEFT is more effective for Mamba than Transformers.

Modified PEFT methods improve Mamba adaptation.

Combining multiple PEFT methods yields superior performance.

Abstract

An ecosystem of Transformer-based models has been established by building large models with extensive data. Parameter-efficient fine-tuning (PEFT) is a crucial technology for deploying these models to downstream tasks with minimal cost while achieving effective performance. Recently, Mamba, a State Space Model (SSM)-based model, has attracted attention as a potential alternative to Transformers. While many large-scale Mamba-based models have been proposed, efficiently adapting pre-trained Mamba-based models to downstream tasks remains unexplored. In this paper, we conduct an exploratory analysis of PEFT methods for Mamba. We investigate the effectiveness of existing PEFT methods for Transformers when applied to Mamba. We also modify these methods to better align with the Mamba architecture. Additionally, we propose new Mamba-specific PEFT methods that leverage the distinctive structure of Mamba. Our experiments indicate that PEFT performs more effectively for Mamba than Transformers. Lastly, we demonstrate how to effectively combine multiple PEFT methods and provide a framework that outperforms previous works. To ensure reproducibility, we will release the code after publication.