search

DLR-SC / corpus-annotation-graph-builder

Corpus Annotation Graph builder (CAG) is an architectural framework that employs the build-and-annotate pattern for creating a graph.
MIT License
10 stars 1 forks source link

Using CAG ArangoDB Graph To incorporate with LLM #46

Open roxanneelbaff opened 11 months ago

roxanneelbaff commented 11 months ago

We want to use our Arango graph to be incorporated with LLM.

As input, the new model will take a graph/subgraph, process it, save it in the vector db and create a RAG.

CONFIG

For that, we define all the attributes in config file (YAML or python dictionary) with the following variables (preliminary):

model_name_path: "llama-v2-b-chat-hf"

generation_params:
          temperature: 0.1 
          ....

 graph_db:
          username: 
          password:
          url:....   

vector_db:
          type: chroma_db
          username: 
          password:
          url:.... 

RAG_params: <params related to integrating vector db to llm>

CagRag Model

The CAG RAG model takes the config file and the subgraph (using AQL output, or a postprocessed format) and outputs the RAG model.

Questions to answer