Don't Start Over: A Cost-Effective Framework for Migrating Personalized Prompts Between LLMs

2026-01-17

AI Summary

This work addresses the challenge that updates to large language models often render user-customized soft prompts obsolete, necessitating costly full retraining. To overcome this, we propose PUMA, a lightweight framework that enables efficient transfer of personalized prompts across incompatible large language model architectures for the first time. PUMA integrates parameter-efficient fine-tuning adapters with a grouped user selection strategy, supporting complex migration scenarios such as chaining and aggregation. Extensive experiments on three large-scale datasets demonstrate that PUMA achieves performance comparable to or even surpassing that of training from scratch, while reducing computational costs by up to 98%. The framework also exhibits strong generalization and robustness across diverse settings.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Prompt Engineering / Prompting
• Planning, Routing, and Scheduling: Planning with Language Models

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Search and Retrieval-Augmented AI: Large language models for search
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

Personalization in Large Language Models (LLMs) often relies on user-specific soft prompts. However, these prompts become obsolete when the foundation model is upgraded, necessitating costly, full-scale retraining. To overcome this limitation, we propose the Prompt-level User Migration Adapter (PUMA), a lightweight framework to efficiently migrate personalized prompts across incompatible models. PUMA utilizes a parameter-efficient adapter to bridge the semantic gap, combined with a group-based user selection strategy to significantly reduce training costs. Experiments on three large-scale datasets show our method matches or even surpasses the performance of retraining from scratch, reducing computational cost by up to 98%. The framework demonstrates strong generalization across diverse model architectures and robustness in advanced scenarios like chained and aggregated migrations, offering a practical path for the sustainable evolution of personalized AI by decoupling user assets from the underlying models.

Problem

• personalized prompts
• model migration
• large language models
• soft prompts
• user personalization

Innovation

• prompt migration
• parameter-efficient adaptation
• personalized LLMs
• user-specific soft prompts
• model upgrading