Leveraging Self-Attention for Input-Dependent Soft Prompting in LLMs

Ananth Muppidi; Abhilash Nandy; Sambaran Bandyopadhyay

返回文献探索

Paper · arXiv 2506.05629

Leveraging Self-Attention for Input-Dependent Soft Prompting in LLMs

Ananth Muppidi, Abhilash Nandy, Sambaran Bandyopadhyay

37 upvotesJune 5, 2025arXiv 预印本

AI 摘要

A new method using input-dependent soft prompting with a self-attention mechanism improves parameter-efficient fine-tuning for large language models, enhancing zero-shot domain transfer.

soft promptingparameter-efficient fine-tuningpre-trained modelsdownstream tasksInput Dependent Soft Prompting techniqueself-Attention Mechanismzero shot domain transfer

Abstract

The performance of large language models in domain-specific tasks necessitates fine-tuning, which is computationally expensive and technically challenging. This paper focuses on parameter-efficient fine-tuning using soft prompting, a promising approach that adapts pre-trained models to downstream tasks by learning a small set of parameters. We propose a novel Input Dependent Soft Prompting technique with a self-Attention Mechanism (ID-SPAM) that generates soft prompts based on the input tokens and attends different tokens with varying importance. Our method is simple and efficient, keeping the number of trainable parameters small. We show the merits of the proposed approach compared to state-of-the-art techniques on various tasks and show the improved zero shot domain transfer capability.