Graph RAG

Setup

Use Python 3.12

Run ./setup.sh

Copy .env.sample to .env and populate with Anthropic key or Watsonx keys and Astra keys.

There are different input files to test:

• ./state_of_union_2022_clip.txt
• ./state_of_union_2022_full.txt
• ./tech_report_2024.txt

Then you have different question sets you can use or change

# Questions
questions1 = [
    "How is AI used across different industries?",
    "What are all the climate technologies mentioned?",
    "What happened after quantum breakthroughs?" 
]

questions2 = [
    "What did Biden say about Ukraine?",
    "What did Biden say about inflation?",
    "What did Biden mention about COVID-19?"
]

Main demo scripts

• vanilla-llm.py simple use of Anthropic Claude LLM to answers questions
• vanilla-vs-rag.py use a RAG approach, holding chunks and documents in memory and compare answers to questions against vanilla LLM
• wx-simple-rag-vs-graph.py use IBM watsonx.ai and its open source models; storing in a vector database, compare simple upload vs knowledge grpah upload

IBM watsonx.ai setup

https://dataplatform.cloud.ibm.com/wx/home?context=wx

Projects > New Project > - Name: graph-rag - Select object storage: my-object-storage

Project graph-rag

• Manage > Service Integrations > Associate Service

  • Watson Machine Learning-1f3 (find the instance of machine learning engine)
    • Plan: Essentials
    • Location: Dallas

• Manage > General Project id: [get the project id] (WATSONX_PROJECT_ID)

Get an API key (WATSONX_API_KEY)

• IBM Cloud - Manage IAM > API Key