This paper presents a comprehensive survey of efficient prompting methods for large language models (LLMs). Prompting has become a mainstream approach for adapting LLMs to specific natural language processing tasks. However, traditional prompting methods require significant computational resources and manual effort, especially for long and complex prompts. To address these challenges, efficient prompting methods have been developed, which can be broadly categorized into two approaches: prompting with efficient computation and prompting with efficient design. Prompting with efficient computation focuses on reducing the computational cost of prompts through techniques such as prompt compression. This includes knowledge distillation, encoding, and filtering. Knowledge distillation involves training a smaller model to mimic the behavior of a larger model, while encoding compresses prompts into vector representations. Filtering removes redundant information from prompts to improve efficiency. Prompting with efficient design aims to automate the design of prompts using techniques such as gradient-based optimization and evolution-based algorithms. Gradient-based methods use optimization algorithms to find the best prompts, while evolution-based methods use genetic algorithms to search for optimal prompts. The paper also discusses the challenges of prompt design, including the need for efficient prompt compression and the difficulty of designing prompts that are effective for LLMs. It highlights the importance of efficient prompting methods in reducing the computational and financial costs associated with LLMs. The survey concludes that efficient prompting methods are essential for the effective use of LLMs in both academic and commercial settings. Future research directions include further exploration of efficient prompting methods and their theoretical foundations. The paper provides a detailed overview of the current state of efficient prompting methods and their potential for future development.This paper presents a comprehensive survey of efficient prompting methods for large language models (LLMs). Prompting has become a mainstream approach for adapting LLMs to specific natural language processing tasks. However, traditional prompting methods require significant computational resources and manual effort, especially for long and complex prompts. To address these challenges, efficient prompting methods have been developed, which can be broadly categorized into two approaches: prompting with efficient computation and prompting with efficient design. Prompting with efficient computation focuses on reducing the computational cost of prompts through techniques such as prompt compression. This includes knowledge distillation, encoding, and filtering. Knowledge distillation involves training a smaller model to mimic the behavior of a larger model, while encoding compresses prompts into vector representations. Filtering removes redundant information from prompts to improve efficiency. Prompting with efficient design aims to automate the design of prompts using techniques such as gradient-based optimization and evolution-based algorithms. Gradient-based methods use optimization algorithms to find the best prompts, while evolution-based methods use genetic algorithms to search for optimal prompts. The paper also discusses the challenges of prompt design, including the need for efficient prompt compression and the difficulty of designing prompts that are effective for LLMs. It highlights the importance of efficient prompting methods in reducing the computational and financial costs associated with LLMs. The survey concludes that efficient prompting methods are essential for the effective use of LLMs in both academic and commercial settings. Future research directions include further exploration of efficient prompting methods and their theoretical foundations. The paper provides a detailed overview of the current state of efficient prompting methods and their potential for future development.