CLIPer: Tailoring Diverse User Preference via Classifier-Guided Inference-Time Personalization

2026-05-07

AI Summary

This work addresses the high computational cost and limited scalability of conventional personalized large language models, which typically rely on extensive fine-tuning to accommodate diverse user preferences. The authors propose a lightweight, inference-time personalization approach that introduces, for the first time, a classifier-guided mechanism during generation. By dynamically modulating the decoding process through real-time fusion of preference classification signals, the method enables personalized text generation without any additional fine-tuning. It supports both single- and multi-dimensional preference control and consistently produces high-quality, tailored outputs across various preference dimensions. This approach significantly reduces computational overhead while offering strong controllability and excellent scalability.

Abstract

Personalized LLMs can significantly enhance user experiences by tailoring responses to preferences such as helpfulness, conciseness, and humor. However, fine-tuning models to address all possible combinations of user preferences is computationally expensive and impractical. In this paper, we introduce \textbf{CLIPer}(\textbf{Cl}assifier-guided \textbf{I}nference-time \textbf{Per}sonalization), a lightweight personalization approach that leverages a classifier model to steer LLM generation dynamically to different user preferences at inference time. Our method eliminates the need for extensive fine-tuning, inducing negligible additional computational overhead while enabling more controllable and nuanced personalization across single and multi-dimensional preferences. Comprehensive empirical analyses demonstrate the scalability and effectiveness of our approach in delivering personalized language generation.

Problem

• personalization
• large language models
• user preferences
• inference-time adaptation
• computational efficiency

Innovation

• inference-time personalization
• classifier-guided generation
• personalized LLMs
• lightweight personalization
• controllable text generation