WiP: Towards Light Adaptation of Large Language Models For Personal Hardware

This work investigates techniques for adapting large language models to run efficiently on personal hardware with limited computational resources and memory.

We explore lightweight adaptation methods that enable the deployment of modern LLMs on consumer devices while preserving model performance. Our approach focuses on reducing hardware requirements through efficient fine-tuning strategies, addressing the growing need for personalized AI assistants that can operate locally without requiring constant cloud connectivity.

Download paper here

Recommended citation: Liangyu Wang, Junxiao Wang and Di Wang. (2024). "WiP: Towards Light Adaptation of Large Language Models For Personal Hardware." Mobisys workshop 2024.
Download Paper