MAPO: Boosting Large Language Model Performance with Model-Adaptive Prompt Optimization This paper introduces MAPO, a model-adaptive prompt optimization method for large language models (LLMs) to enhance their performance on various downstream tasks. The authors demonstrate that different prompts should be adapted to different LLMs to improve their capabilities. MAPO optimizes original prompts for each specific LLM in downstream tasks, leading to significant improvements in performance. The method involves creating a warm-up dataset to generate candidate prompts, followed by supervised fine-tuning (SFT) and reinforcement learning (RL) to refine prompts. The framework also incorporates Proximal Policy Optimization (PPO) and RRMF to further improve performance. Extensive experiments show that MAPO outperforms existing prompt optimization methods, particularly in tasks such as question-answering, classification, and generation. The results indicate that MAPO is effective in enhancing the performance of various LLMs and is particularly useful in low-resource scenarios. The method is evaluated on three popular LLMs (BLOOM, GPT-J, and LLaMA) across different downstream tasks, demonstrating its robustness and generalization. The study also highlights the importance of SFT and RL in prompt optimization and shows that MAPO achieves superior performance compared to other methods. The paper concludes that MAPO is a significant contribution to improving the performance and accuracy of LLMs in practical applications.MAPO: Boosting Large Language Model Performance with Model-Adaptive Prompt Optimization This paper introduces MAPO, a model-adaptive prompt optimization method for large language models (LLMs) to enhance their performance on various downstream tasks. The authors demonstrate that different prompts should be adapted to different LLMs to improve their capabilities. MAPO optimizes original prompts for each specific LLM in downstream tasks, leading to significant improvements in performance. The method involves creating a warm-up dataset to generate candidate prompts, followed by supervised fine-tuning (SFT) and reinforcement learning (RL) to refine prompts. The framework also incorporates Proximal Policy Optimization (PPO) and RRMF to further improve performance. Extensive experiments show that MAPO outperforms existing prompt optimization methods, particularly in tasks such as question-answering, classification, and generation. The results indicate that MAPO is effective in enhancing the performance of various LLMs and is particularly useful in low-resource scenarios. The method is evaluated on three popular LLMs (BLOOM, GPT-J, and LLaMA) across different downstream tasks, demonstrating its robustness and generalization. The study also highlights the importance of SFT and RL in prompt optimization and shows that MAPO achieves superior performance compared to other methods. The paper concludes that MAPO is a significant contribution to improving the performance and accuracy of LLMs in practical applications.