CRAFT: prompts optimized for accuracy-cost Pareto front

Arxiv CS.CL (NLP/language models)

CRAFT searches the Pareto front of prompt accuracy vs. token cost, aiming for optimal trade-off per task and budget. The method refines prompts to reduce inference costs while maintaining accuracy.

prompt-engineeringllm

Arxiv CS.CL (NLP/language models)·Shanu KumarShubhanshu KhandelwalAkhila Yesantarao VenkataParag AgrawalYova KementchedjhievaManish Gupta

7 days ago