Reproduce “Multi-Teacher Reasoning Distillation for LLM Recommendation with Self-Exploration” 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

Liangbo Ning, Siyuan Liang, Yi Zhou, Yifeng Zhang, Qing Li, Wenqi Fan, Dacheng Tao

MUSE trains a recommender to use preference reasoning selected for its effect on ranking the known next item. It gathers candidate summaries from multiple teachers and progressively from the student, then uses rank improvement to choose preferred and dispreferred examples for training the generator and recommendation encoder. On Video Games, Qwen2.5-3B reaches H@10 of 0.1020 versus R²ec’s 0.0921, one benchmark result.

Summary by alphaXiv.

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