Prompt Tuning via Self-Generated Evaluation Questions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing prompt tuning methods for large language models are resource-intensive and time-consuming, requiring significant computing resources and user-provided input/output pairs, which is impractical for individual users with varied tasks.
Innovation Solution
A prompt tuning method that involves guiding a first model to start a question answering process based on the original prompt, using an agent module to interact with the model, and generating optimized prompts through a series of interactions, reducing the need for user-provided data and computational resources.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional prompt tuning methods using input/output pairs are used, then candidate prompts can be generated and evaluated, but computing resources and time are occupied excessively
Solution Approach 1:
The patent introduces an agent module as an intermediary between the user and the large model. This agent automatically generates questions based on the original prompt and obtains answers, creating a mediation layer that eliminates the need for manual input/output pair creation while maintaining effective prompt evaluation capability
Solution Approach 2:
The system enables self-service by allowing the large model to automatically generate its own evaluation data. The model questions itself based on the original prompt and answers these questions, creating a self-contained evaluation mechanism that does not require external human input or extensive computational resources
2Measurement precision
If conventional prompt tuning methods requiring user-provided input/output pairs are used, then prompt optimization can be achieved, but user burden and data requirements increase
Solution Approach 1:
The system allows the large model to serve itself by automatically generating evaluation questions and answers based on the original prompt. This self-service mechanism eliminates the need for users to provide input/output pairs, making the system accessible to individual users without requiring them to prepare training data
Solution Approach 2:
The agent module acts as an intermediary that handles all data generation and evaluation tasks automatically. It translates the original prompt into evaluation questions and collects answers, shielding users from the complexity of data preparation while maintaining high optimization quality
Data Source
AI summary
A prompt tuning method, a prompt tuning apparatus, and a non-transitory computer-readable recording medium are provided. In the method, an original prompt input by a user is received. Then, a first model is guided to start a question answering process based on the original prompt, and the first model is requested to answer the original prompt according to the original prompt and context information obtained in the question answering process. The question answering process includes at least one of a process where the first model asks a question and the second model answers the question, and a process where the first model asks a question and answers the question. Then, the answer to the original prompt generated by the first model is obtained.