MemTrace: Tracing and Attributing Errors in Large Language Model Memory Systems

arXiv cs.CL · 6 hours ago

Researchers developed MemTrace, a framework for tracing and attributing errors in large language model memory systems by converting memory pipelines into executable graphs. The method identified systematic failure modes in memory systems like RAG and Mem0, with automatic attribution pinpointing root causes of failures such as information loss and retrieval misalignment. Using these attribution signals to optimize prompts improved end-task performance by up to 7.62% in a closed-loop correction system.