Large Model Sample Generation for Low-Label Ranking Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing large language models require extensive manual labeling for fine-tuning, leading to high labor and time costs, which are not sustainable for the increasing volume of data in business applications like finance and market research.
Innovation Solution
A method using a large language model to generate candidate questions and indicators from an indicator database, reducing reliance on manual labeling by generating and recalling indicators based on example samples, and training a ranking model for efficient sample generation and labeling.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual labeling is used for fine-tuning large language models, then model accuracy can be improved, but labor and time costs increase significantly
Solution Approach 1:
The patent uses a large language model to automatically generate synthetic labeled samples by copying and transforming existing data patterns, replacing manual labeling. The model generates candidate questions, indicators, and answers based on learned patterns from training data, creating synthetic samples that mimic real labeled data without requiring human annotators for each sample.
Solution Approach 2:
The system implements self-service by enabling the large language model to perform self-labeling of training data. The model automatically generates labels, questions, and indicators for training samples without external human intervention, allowing it to self-improve through iterative training on synthetically generated data.
2Measurement precision
If manual labeling is used for fine-tuning large language models, then model accuracy can be improved, but labor costs increase significantly
Solution Approach 1:
The patent uses a large language model to automatically generate synthetic labeled samples by copying and transforming existing data patterns, replacing manual labeling. The model generates candidate questions, indicators, and answers based on learned patterns from training data, creating synthetic samples that mimic real labeled data without requiring human annotators for each sample.
Solution Approach 2:
The system implements self-service by enabling the large language model to perform self-labeling of training data. The model automatically generates labels, questions, and indicators for training samples without external human intervention, allowing it to self-improve through iterative training on synthetically generated data.
3Reliability
If extensive manual labeling is performed, then model performance improves, but processing efficiency decreases
Solution Approach 1:
The patent uses a large language model to automatically generate synthetic labeled samples by copying and transforming existing data patterns, replacing manual labeling. The model generates candidate questions, indicators, and answers based on learned patterns from training data, creating synthetic samples that mimic real labeled data without requiring human annotators for each sample.
Solution Approach 2:
The system changes the parameter of data volume by generating large quantities of synthetic samples through automated processes. The large language model can produce extensive training data rapidly, transforming the parameter of sample quantity from a manual bottleneck to an scalable automated output, thereby improving processing efficiency while maintaining model performance.
Data Source
AI summary
A large model-based method of generating a sample, a method of training a model, a ranking method, and a device are provided, which relate to a field of artificial intelligence technology, and in particular to fields of intelligent search, deep learning, natural language processing and large model technologies. The method includes: determining indicators from initial indicators contained in an indicator database in response to a sample generation request, where the sample generation request contains an example sample; generating candidate questions based on the indicators by using a question example contained in the example sample as a basic corpus; recalling candidate indicators corresponding to each candidate question from the initial indicators; and generating target samples based on the candidate questions and the candidate indicators corresponding to each candidate question by using example samples as the basic corpus.