Reproduce “Hybrid Generative-Tagging Decoding for Efficient Product Attribute-Value Extraction” 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

Keshav Kolluru, Manas Gupta, Maroof Ahmad, Suchitra Sathyanarayana

The authors introduce a hybrid system for extracting product attributes from catalog text. It uses one language-model prefill to copy contiguous values with a span tagger and generate implicit or non-copyable values. On a text-only Amazon validation task, it achieves 80.2% recall versus 69.8% for generation, equivalent judged precision, and lower latency. Mean per-product time is 31.5 versus 56.6 milliseconds in the tested serving setup; findings are limited to this setting.

Summary by alphaXiv.

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