NPO-optimized prompts transfer well to other student models, especially within the same model family

NPO-optimized prompts elicit similar performance improvements when applied verbatim to other student models, especially across models within the same family

http://arxiv.org/abs/2608.27266v1