State-offset Tuning: State-based Parameter-Efficient Fine-Tuning for State Space Models
Wonjun Kang, Kevin Galim, Yuchen Zeng, Minjae Lee, Hyung Il Koo, Nam Ik Cho
2025-03-11
Summary
This paper talks about a new way to fine-tune AI models that handle time-based data (like audio or sensor readings) by tweaking their internal memory instead of adding external hints, making them more efficient.
What's the problem?
Current methods for adapting these models to new tasks (like speech recognition) use external prompts that don’t work well with their structure, leading to slower and less effective learning.
What's the solution?
The new method, State-offset Tuning, adjusts how the model’s internal memory tracks information over time, letting it learn new tasks faster and with fewer changes to its original setup.
Why it matters?
This makes AI tools for real-time tasks (like voice assistants or health monitoring) cheaper and easier to adapt to new uses without losing speed or accuracy.
Abstract
State Space Models (SSMs) have emerged as efficient alternatives to Transformers, mitigating their quadratic computational cost. However, the application of Parameter-Efficient Fine-Tuning (PEFT) methods to SSMs remains largely unexplored. In particular, prompt-based methods like Prompt Tuning and Prefix-Tuning, which are widely used in Transformers, do not perform well on SSMs. To address this, we propose state-based methods as a superior alternative to prompt-based methods. This new family of methods naturally stems from the architectural characteristics of SSMs. State-based methods adjust state-related features directly instead of depending on external prompts. Furthermore, we introduce a novel state-based PEFT method: State-offset Tuning. At every timestep, our method directly affects the state at the current step, leading to more effective adaptation. Through extensive experiments across diverse datasets, we demonstrate the effectiveness of our method. Code is available at https://github.com/furiosa-ai/ssm-state-tuning.