Prompt Vector Determination via Pruned Model Fusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The accuracy of prompt vectors determined by pre-trained models in natural language processing is inadequate due to reliance on single models, leading to one-sided and inaccurate results.
Innovation Solution
A method involving the fusion of sample data vectors with prompt vectors and inputting them into multiple pruned models to obtain scores, iteratively modifying the prompt vectors until a target prompt vector is determined, allowing for comprehensive and reliable analysis through multiple perspectives.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single pre-trained model is used to determine prompt vectors, then the process is simple and fast, but the accuracy and comprehensiveness of the prompt vector is inadequate
Solution Approach 1:
The patent divides the single pre-trained model into multiple pruned models by segmenting the model parameters. Each pruned model is trained on different subsets of data or with different parameter selections, allowing the system to evaluate prompt vectors from multiple perspectives simultaneously, thereby improving accuracy without requiring a completely new architectural approach
Solution Approach 2:
The patent merges the results from multiple pruned models to determine the final prompt vector. By combining the outputs or evaluations from N different pruned models, the system achieves more comprehensive and accurate prompt vector determination, resolving the contradiction between simplicity and accuracy
2Measurement precision
If multiple pruned models are used to evaluate prompt vectors, then the accuracy and comprehensiveness improve, but the computational resources and processing time increase
Solution Approach 1:
The patent performs preliminary actions by pre-pruning the model into N different pruned models before actual prompt vector evaluation. This preparation allows the system to efficiently evaluate multiple prompt vectors using the pre-divided model structures, reducing the computational burden during the actual evaluation phase compared to using a single full model repeatedly
Solution Approach 2:
The patent creates multiple copies (pruned models) of the original pre-trained model, each optimized for different aspects of evaluation. These copies can process prompt vectors in parallel, improving throughput while maintaining accuracy, thereby reducing the effective computational resources needed per evaluation
3Measurement precision
If iterative modification of prompt vectors is performed, then the accuracy converges to target level, but the processing time increases
Solution Approach 1:
The patent implements a feedback mechanism where the evaluation results from the N pruned models are used to iteratively modify the prompt vector. The system receives feedback about the quality of the prompt vector from multiple model perspectives and adjusts accordingly, converging to an accurate target prompt vector through systematic iterations
Solution Approach 2:
The patent makes the prompt vector determination process dynamic by allowing iterative modifications based on evaluation feedback. The system adapts the prompt vector dynamically through multiple rounds of evaluation and adjustment, converging to the optimal solution while managing processing time through efficient iteration strategies
Data Source
AI summary
A method for determining a prompt vector of a pre-trained model, includes: obtaining a first one of prompt vectors and a first vector corresponding to sample data; obtaining N pruned models by N different pruning processing on the pre-trained model, where N is any integer greater than 1; obtaining a first score corresponding to the first one of the prompt vectors by fusing the first vector and the first one of the prompt vectors and inputting the fused first vector and first one of the prompt vectors into the N pruned models respectively; determining a second one of the prompt vectors by modifying, based on the first score, the first one of the prompt vectors; and based on the second one of the prompt vectors, returning to obtaining the first score until determining a target prompt vector corresponding to the sample data.