Reproduce “SPRINT: Single-Step Generative Recommendation via Average Probability Velocity” 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
Zhuo Cai, Shoujin Wang, Peilin Zhou, Min Xu, Julian McAuley, Fang Chen
SPRINT generates recommendations by predicting all four-token semantic codes in one forward pass. A bidirectional Transformer estimates average token probabilities from an all-mask code, while a contrastive whole-code score helps keep tokens associated with one item. On Scientific, it reaches NDCG@5 of 0.0301 versus 0.0273 for DiffGRM. Its test-set inference is 8.39 times faster than RPG, not a production guarantee.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.