PromptWizard, the latest open source tool released by the Microsoft AI research team, provides a new solution for large language model (LLM) prompt word optimization. Traditional prompt word optimization methods are time-consuming and difficult to expand. PromptWizard significantly improves model performance and efficiency by introducing a feedback mechanism and adopting a critical and comprehensive iterative optimization strategy. It uses LLM to generate and evaluate multiple prompt word variants in the generation phase, ensuring continuous performance improvement in the test inference phase, and ultimately achieves excellent results on multiple data sets and significantly reduces resource consumption.
Recently, the Microsoft AI research team released the open source tool PromptWizard, a feedback-driven AI framework designed to efficiently optimize the prompt design of large language models (LLM). The quality of hints is critical to the quality of model output, however, creating high-quality hints often requires a lot of time and human resources, especially in complex or domain-specific tasks.
Traditional prompt optimization methods mostly rely on manual experience, which is not only time-consuming but also difficult to expand. Existing optimization techniques are divided into two types: continuous and discrete. Continuous techniques such as soft prompts require extensive computing resources, while discrete methods such as PromptBreeder and EvoPrompt are evaluated by generating multiple prompt variants. Although these methods perform well in some cases, they lack an effective feedback mechanism, which often results in The results were unsatisfactory.
PromptWizard significantly improves task performance by introducing a feedback mechanism and using a critical and comprehensive approach to repeatedly optimize prompt instructions and examples. Its workflow is mainly divided into two stages: the generation stage and the test inference stage. In the generation phase, the system leverages a large language model to generate multiple variants of the underlying cues and evaluates them to find high-performing candidates. At the same time, the framework's built-in criticism mechanism will analyze the advantages and disadvantages of each prompt and provide feedback to guide subsequent optimization. After multiple rounds of optimization, the system can improve the diversity and quality of prompts.
During the test inference phase, optimized hints and examples are applied to new tasks to ensure continued performance improvement. With this approach, PromptWizard conducts extensive experiments on 45 tasks and achieves excellent results in both unsupervised and supervised settings. For example, it achieves 90% unsupervised accuracy on the GSM8K dataset and 82.3% on SVAMP. In addition, compared to discrete methods such as PromptBreeder, PromptWizard reduces API calls and token usage by up to 60 times, demonstrating its efficiency in resource-constrained environments.
PromptWizard's success lies in its innovative sequence optimization, guided critique, and integration of expert roles, allowing it to be effectively adapted to specific tasks and have good interpretability. This progress heralds the importance of automation frameworks in natural language processing workflows and is expected to promote more effective and economical applications of advanced AI technologies.
Blog: https://www.microsoft.com/en-us/research/blog/promptwizard-the-future-of-prompt-optimization-through-feedback-driven-self-evolving-prompts/
Project code: https://github.com/microsoft/PromptWizard?tab=readme-ov-file
Paper: https://www.microsoft.com/en-us/research/publication/promptwizard-task-aware-agent-driven-prompt-optimization-framework/
Highlight:
PromptWizard is a new AI framework used to optimize prompts for large language models and improve model performance.
This framework combines critique mechanisms and feedback loops to efficiently generate and evaluate multiple prompt variants.
PromptWizard shows excellent accuracy in multiple tasks and significantly reduces resource consumption and cost.
All in all, PromptWizard provides a powerful tool for prompt word optimization of large language models through innovative feedback-driven mechanisms and efficient optimization strategies. Its efficiency and accuracy give it significant advantages in practical applications, providing a powerful tool for AI Contribute to the development of technology. Interested readers can visit the links provided for more information.