PromptBridge framework translates prompts between models using MAP‑RPE without retraining

PromptBridge is a zero‑training framework that translates optimized prompts from a source model to an effective prompt for a target model. It first calibrates using a small set of alignment tasks via MAP‑RPE to produce paired best prompts, then learns a mapping that generalizes to novel tasks. Claims: low calibration cost, no LLM finetuning, plug‑and‑play transfer across single and multi‑agent scenarios. Paper: https://t.co/ODlps4fqkd