Search papers, labs, and topics across Lattice.
University of Maryland, College Park
3
0
6
5
Adjusting the KL penalty in self-distillation can transform a brittle method into a robust framework that significantly boosts reasoning performance.
Ditch the assumption of tutor correctness: GATES leverages agreement between multiple reasoning traces to distill knowledge in document-grounded QA, boosting accuracy by over 15% on in-domain tasks.
Skip the slow diffusion grind: this generative model matches diffusion model quality in a single pass by encoding images onto a sphere.
