Reproduce “KVUse: Reusable-KV User Memory for Efficient LLM Recommenders” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Xin Guo, Qiang Zhang, Xiang Zhou, Arushi Rai, Siyu Lin, Yifan Wu, Xiang Shen, Lei Huang and 3 more

KVUse reduces repeated prefill in language-model recommenders by storing reusable key-value states for user profiles and item descriptions. A LoRA writer creates profile states during updates, which a frozen reader reuses for scoring. Across four tested datasets, ranking differences from full prefill were not statistically significant; Amazon Books scoring throughput rose 3.96× under saturated load, in a single-backbone leave-one-out evaluation.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.