Search papers, labs, and topics across Lattice.
Affiliation:
3
0
6
Hi-Q outperforms traditional multi-hop QA methods by dynamically refining queries based on evidence support, achieving up to 15.1 EM points higher than the best iterative retrieval baseline.
By dynamically balancing fast adaptation and stable averaging, AMUSE delivers faster convergence and better final performance than AdamW and Muon, all without any learning rate tuning.
Current QA models boasting 70+ F1 on existing benchmarks crumble on SPARTA, a new dataset revealing their surprisingly shallow reasoning abilities across tables and text.
