Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

PLMM: Personal Large Language Models on Mobile Devices

Yuanhao Gong

TL;DR

This work targets privacy and scalability challenges in large language models by proposing Personal Large Language Models (PLMM) that run on local devices and adapt to users' backgrounds and interests. It introduces a three-level P-E-T architecture with P-models on-device, E-models on servers, and T-models for broad knowledge, enabling real-time personalized interactions while preserving privacy. An economic framework accompanies the architecture, wherein users pay for services and may be rewarded for contributions, and developers monetize while funding or compensating user input, enabling a sustainable ecosystem. The approach aims to balance privacy, efficiency, and accuracy with broad applicability to language and vision tasks.

Abstract

Inspired by Federated Learning, in this paper, we propose personal large models that are distilled from traditional large language models but more adaptive to local users' personal information such as education background and hobbies. We classify the large language models into three levels: the personal level, expert level and traditional level. The personal level models are adaptive to users' personal information. They encrypt the users' input and protect their privacy. The expert level models focus on merging specific knowledge such as finance, IT and art. The traditional models focus on the universal knowledge discovery and upgrading the expert models. In such classifications, the personal models directly interact with the user. For the whole system, the personal models have users' (encrypted) personal information. Moreover, such models must be small enough to be performed on personal computers or mobile devices. Finally, they also have to response in real-time for better user experience and produce high quality results. The proposed personal large models can be applied in a wide range of applications such as language and vision tasks.

PLMM: Personal Large Language Models on Mobile Devices

TL;DR

This work targets privacy and scalability challenges in large language models by proposing Personal Large Language Models (PLMM) that run on local devices and adapt to users' backgrounds and interests. It introduces a three-level P-E-T architecture with P-models on-device, E-models on servers, and T-models for broad knowledge, enabling real-time personalized interactions while preserving privacy. An economic framework accompanies the architecture, wherein users pay for services and may be rewarded for contributions, and developers monetize while funding or compensating user input, enabling a sustainable ecosystem. The approach aims to balance privacy, efficiency, and accuracy with broad applicability to language and vision tasks.

Abstract

Inspired by Federated Learning, in this paper, we propose personal large models that are distilled from traditional large language models but more adaptive to local users' personal information such as education background and hobbies. We classify the large language models into three levels: the personal level, expert level and traditional level. The personal level models are adaptive to users' personal information. They encrypt the users' input and protect their privacy. The expert level models focus on merging specific knowledge such as finance, IT and art. The traditional models focus on the universal knowledge discovery and upgrading the expert models. In such classifications, the personal models directly interact with the user. For the whole system, the personal models have users' (encrypted) personal information. Moreover, such models must be small enough to be performed on personal computers or mobile devices. Finally, they also have to response in real-time for better user experience and produce high quality results. The proposed personal large models can be applied in a wide range of applications such as language and vision tasks.
Paper Structure (11 sections, 1 figure, 1 table)

This paper contains 11 sections, 1 figure, 1 table.

Table of Contents

  1. Introduction
  2. Limitation of Large Models
  3. Motivation and Contributions
  4. Personal Large Models
  5. P-E-T Models
  6. P-model
  7. Interaction
  8. Economics
  9. Users
  10. Developers
  11. Conclusion and Discussion

Figures (1)

  • Figure 1: The proposed personal large models run on local machines such as mobile phones and personal computers. Their parameters are trained by the user's personal input and also updated by the expert level models. The expert and traditional level large language models run on servers or cloud. The expert level models are good at specific fields, such as medical image analysis, finance, python coding, etc. The traditional models are good at merging the knowledge from all expert models and updating the expert models. This architecture is abbreviated as P-E-T for convenience and each level model is called P-model, E-model and T-model, respectively.