Inspired by a Gartner webinar. Given the lower energy requirements and the one-shot learning capabilities of functor models the LLM Offloading Pattern is quite useful. It suggests an alternative to scaling for the LLM in certain cases. Here agents work with functor micro models to discover new data. The data processor handles curation: cleansing, dedups, canonicalization, learning preparation, etc. This product of Data Science can be essential in many cases such as fraud detection where unsupervised learning has more applicability.

​

Then the learning router submits the learning unit to the appropriate SLM. The LLM can invoke inference on any SLM to gain knowledge, generate code, etc. The idea when this is used as a knowledge base is to simplify the LLMs task to only the higher levels of taxonomy in the knowledge base. The LLM can contain a map of which SLM contains information on which topic and can direct the request to the correct SLM. The Graph Neural Networks are contained in a new type of ensemble that allows messaging between nodes in different models using an orchestrator. See this document and the diagram below for further details.