Reproduce “Contextual Embeddings Beyond the Gold Chunk” 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
Sedigheh Eslami, Markus Krimmel, William Barber, Antoine Chaffin, Maximilian Schall, Jonas Golde, Louis Milliken, Ekaterina Glazkova and 2 more
The authors present a retrieval method that trains contextual chunk embeddings to retrieve both answer-bearing text and supporting document context. A query-aware compression teacher provides graded token relevance, which is aggregated for different chunkings, alongside document-level contrastive training. On Q2D-Chunk, the 9B model reaches 86.4% Answer@10 versus 83.9% for Voyage-context-4, while Nemotron leads evidence recall; evidence retrieval remains challenging.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.