Language Model Training Progress Detection by Answer Similarity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing language models struggle with determining the progress of training, leading to insufficient or excessive data acquisition, which affects their ability to accurately interpret domain-specific expressions and increases training costs.
Innovation Solution
An information processing apparatus and method that automatically determines the progress of training by comparing answer sentences generated before and after updating the language model using training data, assessing similarity between these sentences to identify effective learning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the number of learning target data is increased to improve training quality, then the accuracy of interpreting domain-specific expressions is improved, but the training cost increases
Solution Approach 1:
The patent implements automatic determination of training progress by comparing answer sentences generated before and after training updates. This feedback mechanism allows the system to monitor whether training is improving model performance on domain-specific expressions, enabling timely termination when sufficient accuracy is achieved, thus avoiding excessive training costs while ensuring adequate learning quality
2Reliability
If training continues until sufficient accuracy is achieved, then the quality of domain-specific word interpretation is improved, but the training time increases
Solution Approach 1:
The system automatically evaluates training progress by comparing model outputs before and after each training update. This feedback loop enables real-time monitoring of learning effectiveness, allowing the training process to terminate when sufficient accuracy on domain-specific expressions is achieved, thereby avoiding unnecessary training time while ensuring quality
Solution Approach 2:
The patent performs preliminary comparison of answer sentences before and after training updates to predict whether training is effective. This preliminary evaluation allows the system to make informed decisions about continuing or terminating training, optimizing the balance between training time and quality without needing to wait for extensive training to confirm effectiveness
Data Source
AI summary
An information processing apparatus includes: at least one memory storing instructions; and at least one processor configured to execute the instructions to; input a first question sentence regarding a content of a target document to a language model updated by learning the content of the target document to generate a first answer sentence; and determine similarity between the first answer sentence and a second answer sentence generated by inputting the first question sentence to the language model before update. With the information processing apparatus, it is possible to efficiently generate a language model optimized for generating an answer sentence for the target document.