Never Start from Scratch: Expediting On-Device LLM Personalization via Explainable Model Selection

2025-04-15

AI Summary

To address the dual challenges of limited on-device computational resources and sparse user data in mobile LLM personalization, this paper proposes an explainability-guided pre-personalization model selection method. Our approach leverages interpretability signals—such as gradient sensitivity and feature attribution—extracted during lightweight fine-tuning to intelligently select the most suitable candidate model for personalization, thereby eliminating the need for from-scratch training. The method integrates model similarity measurement, efficient fine-tuning evaluation, and edge-aware deployment optimization. Evaluated on mainstream smartphones, it reduces on-device personalization computation cost by 83% and improves data efficiency by 51%. By grounding model selection in interpretable signals, our method significantly enhances both the practicality and transparency of personalized LLMs under resource-constrained conditions.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models
• Search and Optimization: Learning to Search

Application Category

• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

Personalization of Large Language Models (LLMs) is important in practical applications to accommodate the individual needs of different mobile users. Due to data privacy concerns, LLM personalization often needs to be locally done at the user's mobile device, but such on-device personalization is constrained by both the limitation of on-device compute power and insufficiency of user's personal data. In this paper, we address these constraints by fine-tuning an already personalized LLM with user's personal data, and present XPerT, a new technique that ensure proper selection of such already personalized LLMs based on explainability about how they were being fine-tuned. We implemented and evaluated XPerT on various smartphone models with mainstream LLMs, and experiment results show that XPerT reduces the computation costs of on-device LLM personalization by 83%, and improves its data efficiency by 51%.

Problem

• Expediting on-device LLM personalization via model selection
• Reducing computation costs of personalizing LLMs on mobile devices
• Improving data efficiency for on-device LLM fine-tuning

Innovation

• Fine-tuning personalized LLMs with user data
• Explainable model selection for efficiency
• Reduces computation costs by 83%