Reproduce “Route to Remember: Adaptive Memory Selection for LLM Agents” 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
Zeyu Fang, Jun Yin, Yongsheng Mei, Xu Han, Chaitali Joshi, Zhongkai Sun, Jie Hao, Bin Guo and 3 more
The authors train per-user adapters to answer questions about past user interactions from their weights, without memory text in the prompt. On-policy self-distillation transfers answers from a context-aware model, while a hybrid system audits weight-generated answers against retrieved text. On native HaluMem questions, OPSD scores 59.9%, above tested text baselines; retrieval overtakes it after roughly twenty sessions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.