DRESSing Up LLM: Efficient Stylized Question-Answering via Style Subspace Editing

2025-01-24

AI Summary

To address the challenge of efficiently and lightweightly generating stylistically controlled text with large language models (LLMs) in applications such as role-playing, this paper proposes a training-free latent-space representation editing method. The core method explicitly models a disentangled stylistic subspace within intermediate-layer hidden representations of LLMs—identified via singular value decomposition and directional projection—and introduces an adaptive strength-weighting mechanism that dynamically controls the degree of style editing during zero-shot inference, balancing style fidelity and semantic consistency. Evaluated on two newly constructed stylized question-answering benchmarks, the approach achieves a 23.6% improvement in style accuracy over prompt engineering and ITI baselines, while maintaining 98.1% semantic consistency—all without fine-tuning or additional parameters.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Sentiment Analysis, Stylistic Analysis, and Argument Mining
• Computer Vision: Diffusion Models for Vision

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information

Abstract

We introduce DRESS, a novel approach for generating stylized large language model (LLM) responses through representation editing. Existing methods like prompting and fine-tuning are either insufficient for complex style adaptation or computationally expensive, particularly in tasks like NPC creation or character role-playing. Our approach leverages the over-parameterized nature of LLMs to disentangle a style-relevant subspace within the model's representation space to conduct representation editing, ensuring a minimal impact on the original semantics. By applying adaptive editing strengths, we dynamically adjust the steering vectors in the style subspace to maintain both stylistic fidelity and semantic integrity. We develop two stylized QA benchmark datasets to validate the effectiveness of DRESS, and the results demonstrate significant improvements compared to baseline methods such as prompting and ITI. In short, DRESS is a lightweight, train-free solution for enhancing LLMs with flexible and effective style control, making it particularly useful for developing stylized conversational agents. Codes and benchmark datasets are available at https://github.com/ArthurLeoM/DRESS-LLM.

Problem

• Language Models
• Stylistic Transfer
• Efficiency

Innovation

• DRESS
• style transfer
• large language models