Our paper "May the Memory Be With You: Efficient and Infinitely Updatable State for Large Language Models" has been accepted at the 5th ACM Workshop on Machine Learning and Systems (EuroMLSys 2025)!
EuroMLSys is co-located with EuroSys 2025 and will take place on March 31, 2025 in Rotterdam, Netherlands.
The Challenge
Large language models excel at natural language tasks but lack persistent state management for personalized and adaptive interactions. This limits their ability to maintain context across sessions and adapt to individual users.
Our Framework
We propose a framework that endows LLMs with stateful capabilities by combining:
- Retrieval-Augmented Generation (RAG) for efficient knowledge access
- Low-Rank Adaptation (LoRA) for parameter-efficient fine-tuning
Our approach recasts the LLM as an editable component that retains hierarchical knowledge, analogous to deferred merge operations in log-structured merge (LSM) trees.
Key Innovation
The framework enables infinitely updatable state while maintaining efficiency — essential for building personalized AI assistants and adaptive systems.
This work was led by Excel Chukwu (MPI-SWS).