AI RESEARCH

DUET: Optimizing Training Data Mixtures via Feedback from Unseen Evaluation Tasks

arXiv CS.LG

ArXi:2502.00270v3 Announce Type: replace The performance of an LLM depends heavily on the relevance of its