1 paper
Xiaoyu Ma, Yiwen Li, Haoyue Liu +4
Automatic prompt optimization (APO) hinges on the quality of its evaluation signal, yet scoring every prompt candidate on the full training set is prohibitively expensive. Existing…