Reproduce “Residual Trajectory Distillation for Generative Retrieval” 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
Weihao Shen, Wei Chen, Fuwei Zhang, Guojun Liu, Qingsong Hua, Wei Lin, Fuzhen Zhuang
ResTD trains a generative retriever with residual information from a frozen indexer, adding soft supervision for current and future codeword choices while keeping stored item addresses and inference fixed. In a matched English ESCI test, it achieves 14.34 Recall@10 versus 13.56 for future hard-code supervision, a 0.78-point gain; this result is limited to the reported controlled setting.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.