LLM Plugin Invocation via Semantic Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing large language models face challenges in efficiently executing language understanding tasks due to limitations in timeliness of pre-training data, resource dependencies, and high costs associated with plugin extension and customization.
Innovation Solution
A method and apparatus for invoking a plugin of a large language model, which involves acquiring natural language content, performing semantic understanding, detecting plugin hits, determining the to-be-executed session understanding task, and calling the corresponding plugin to obtain the calling result.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If pre-training data is used for large language models, then the model can understand and generate language content, but the timeliness of the data becomes outdated and limits the model's ability to handle current tasks
Solution Approach 1:
The system segments the language understanding process into two parts: a static pre-trained large language model for general capabilities, and a dynamic plugin matching module for real-time task-specific adaptations. This allows the model to maintain its pre-trained reliability while overcoming data timeliness limitations through external plugin resources.
Solution Approach 2:
A plugin matching module is introduced as an intermediary between the pre-trained model and the task execution. This mediator retrieves relevant plugin resources based on the input content, enabling the system to access up-to-date information and capabilities without retraining the entire model, thus resolving the timeliness issue.
2Adaptability or versatility
If plugins are extended and customized for specific tasks, then the functionality and versatility of the system improve, but the cost and complexity of implementation increase
Solution Approach 1:
The system designs a universal plugin interface and matching mechanism that can handle multiple task types through a common framework. The plugin matching module uses semantic analysis to automatically match inputs with appropriate plugins, eliminating the need for complex customizations for each task and reducing overall system complexity while maintaining versatility.
Solution Approach 2:
The plugin matching module automatically retrieves and selects appropriate plugins based on semantic understanding of the input content, without requiring manual configuration or complex customization. This self-service mechanism reduces implementation complexity while maintaining high adaptability to different tasks.
3Measurement precision
If semantic understanding and plugin matching are performed for every input, then the accuracy of task execution improves, but the processing time and computational resources increase
Solution Approach 1:
The system performs semantic understanding and plugin matching selectively rather than for every input. The matching module analyzes the input content and only retrieves plugins when there is a potential match, avoiding unnecessary processing for inputs that don't require plugin execution. This partial action approach maintains accuracy while reducing overall processing time.
Data Source
AI summary
A method for invoking a plugin of a large language model includes: acquiring natural language content; performing semantic understanding on the natural language content and detecting whether the natural language content hits a plugin to obtain a first plugin pointed to by the plugin hit result; comparing the first plugin with a second plugin corresponding to the current session understanding task to determine a to-be-executed session understanding task and a third plugin corresponding to the to-be-executed session understanding task; acquiring the language understanding content of the to-be-executed session understanding task and sending the language understanding content to the large language model to obtain the input parameter of the third plugin; and calling the third plugin according to the input parameter of the third plugin to obtain the calling result of the to-be-executed session understanding task.