Introduced KG which creates a network creates nodes which represents classes, functions or other features of the code. The connecting edges the traits that relates the nodes. According to the query, the LLM traverses across the graph, coming up with answers.
But as a drawback, the ETA for retrieval is quite higher. But by introducing GPU powered machines, this can be better. Hallucination can also be a problem when repo files are smaller.
As an add on, we can introduce Agents, that can route to KG functionality when the user required more accurate and descriptive answers.
![Uploading S
creenshot from 2024-0
7-22 10-49-26.png…]()
Introduced KG which creates a network creates nodes which represents classes, functions or other features of the code. The connecting edges the traits that relates the nodes. According to the query, the LLM traverses across the graph, coming up with answers. But as a drawback, the ETA for retrieval is quite higher. But by introducing GPU powered machines, this can be better. Hallucination can also be a problem when repo files are smaller. As an add on, we can introduce Agents, that can route to KG functionality when the user required more accurate and descriptive answers.
The testing repo I used -
https://github.com/NousResearch/Hermes-Function-Calling