Selective LLM Routing in Multi-Level AI Supercomputers
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current AI generative models and large language models (LLMs) require significant supercomputing efforts to improve response times, accuracy, and reduce computational load, limiting their scalability and expandability.
Innovation Solution
A multi-level generative AI system utilizing derived requests, multiple h-LLMs with varying levels of accuracy, and a distributed architecture to efficiently process and refine tasks, incorporating techniques like bagging, boosting, and merging/fusing h-LLMs to enhance precision and accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If traditional machine learning algorithms are used for content analysis, then computational efficiency is maintained, but the ability to generate human-like content is limited
Solution Approach 1:
The system segments the monolithic LLM into multiple specialized h-LLMs, each trained on specific datasets for particular tasks. This segmentation allows each model to be computationally efficient while the ensemble provides versatile content generation capabilities across different domains.
Solution Approach 2:
The framework creates a universal system where multiple h-LLMs can handle diverse content generation tasks including text, code, and multimedia. The broker architecture enables a single system to perform multiple functions by routing queries to appropriate specialized models.
2Measurement precision
If LLMs are trained on vast amounts of data to improve accuracy, then content generation quality improves, but computational load and training time increase significantly
Solution Approach 1:
Instead of training one large model on all possible data, the system trains multiple smaller h-LLMs on specific local datasets relevant to their specialized tasks. Each model achieves high accuracy in its domain without requiring massive computational resources for general training.
Solution Approach 2:
The broker selectively engages only the necessary h-LLMs for each specific query rather than deploying all models for every task. This partial action approach reduces computational load while maintaining accuracy by using only the relevant specialized models.
3Adaptability or versatility
If a single large LLM is used to handle all tasks, then model versatility is achieved, but response time and computational resources increase
Solution Approach 1:
The system divides the single large LLM into multiple smaller h-LLMs, each optimized for specific task types. This segmentation enables faster response times for individual queries while maintaining overall versatility through the ensemble of specialized models.
Solution Approach 2:
The broker acts as an intermediary that routes queries to the most appropriate h-LLM based on the task requirements. This mediation ensures fast response times by avoiding unnecessary model processing while maintaining versatility through intelligent query routing.
4Measurement precision
If multiple h-LLMs are combined to improve accuracy through bagging and boosting, then precision increases, but system complexity increases
Solution Approach 1:
The broker implements self-service mechanisms by automatically selecting and combining h-LLMs based on query characteristics and model performance metrics. This automation reduces the operational complexity of managing multiple models while maintaining high prediction accuracy through systematic ensemble methods.
Data Source
AI summary
A method for generating a merged large language model (LLM) from input data including receiving derived prompts generated from a user prompt and relevant contexts, receiving the relevant contexts related to the generation of the plurality of derived prompts, providing derived prompts and the relevant contexts to one or more merged LLMs, generating a plurality of results, and sending the plurality of results to an output broker.