MO-CAPO: Multi-Objective Cost-Aware Prompt Optimization
Researchers have developed MO-CAPO, a new algorithm designed to optimize prompts for large language models (LLMs) by considering both performance and inference cost simultaneously. Unlike previous methods that often prioritize only performance, MO-CAPO employs a multi-objective approach that efficiently explores trade-offs between these competing factors. The algorithm aims to provide practitioners with a diverse set of prompts that offer various balances between model accuracy and computational expense, outperforming existing multi-objective baselines in several evaluations. AI
IMPACT Enables more cost-effective deployment of LLMs by optimizing prompts for both performance and inference cost.