GraphAcademy

Build a Neo4j-backed Chatbot using Python

Getting Started

Project Configuration

Building Tools

Creating an LLM Instance

In the Initializing the LLM lesson of Neo4j & LLM Fundamentals, you learned how to initialize an LLM class and generate a response from an LLM.

In this lesson, you will need to put this learning into practice by creating an LLM instance to communicate with a GPT model using OpenAI.

You will need to:

  1. Obtain an API key from platform.openai.com

  2. Create a secrets file to save the API key

  3. Initialize an instance of the ChatOpenAI

  4. Create an instance of the OpenAIEmbeddings model

Setting Streamlit Secrets

To keep it secure, you will store the API key in the Streamlit secrets.toml file.

Create a new file, .streamlit/secrets.toml, and copy the following text, adding your OpenAI API key.

toml
.streamlit/secrets.toml
OPENAI_API_KEY = "sk-..."
OPENAI_MODEL = "gpt-4"

We have found that using gpt-4 yields the best results. There are other models that may work better for your scenario.

You can access values stored in the secrets.toml file using st.secrets:

python
Accessing Secrets
import streamlit as st

openai_api_key = st.secrets['OPENAI_API_KEY']
openai_model = st.secrets['OPENAI_MODEL']

Keep your secrets safe

The Streamlit documentation outlines four approaches to handling secrets and credentials in your application.

Ensure you do not share or include your API keys in a git commit. The .gitignore file includes the .streamlit/secrets.toml file, so git won’t push the API key to Github.

Initializing an OpenAI LLM

As you will use the LLM across the application, you should include the LLM instance in a module that you can import.

Open the llm.py file in the project root.

Create a new llm instance of the ChatOpenAI class:

python
llm.py
# Create the LLM
from langchain_openai import ChatOpenAI

llm = ChatOpenAI(
    openai_api_key=st.secrets["OPENAI_API_KEY"],
    model=st.secrets["OPENAI_MODEL"],
)

The LLM is initialized with the openai_api_key and model stored in the secrets.toml file.

Initializing an Embedding Model

To use the Vector Search Index, you must create an instance of the OpenAIEmbeddings model. Langchain will use this when creating embeddings to find similar documents to the user’s input using Neo4j’s vector index.

python
llm.py
# Create the Embedding model
from langchain_openai import OpenAIEmbeddings

embeddings = OpenAIEmbeddings(
    openai_api_key=st.secrets["OPENAI_API_KEY"]
)

Using the LLM

Once you have completed the steps, you can import the llm and embeddings objects into other modules within the project.

python
from llm import llm, embeddings

That’s it!

Once you have completed the steps above, click the button to mark the lesson as completed.

Summary

In this lesson, you have created the classes required to interact with OpenAI’s LLMs.

In the next lesson, you will create the classes required to connect to Neo4j.