Quickstart Guide for pycolbertdb

This quickstart guide provides instructions on how to use the pycolbertdb package to integrate ColbertDB with LlamaIndex, leveraging OpenAI's GPT-4 model for processing and querying documents.

Prerequisites

Ensure you have the following installed and configured:

• Python 3.x
• An OpenAI API key
• Environment variables configured for ColbertDB

Installation

1. Install the necessary packages

   pip install pycolbertdb -U
   pip install llama-index
   pip install llama-index-readers-web
   pip install requests
   pip install python-dotenv
   

Code Example

Below is an example of how to use the pycolbertdb package to fetch, process, and query documents.

Import Dependencies

Start by importing the necessary dependencies.

import os
from dotenv import load_dotenv
from llama_index.readers.web import SimpleWebPageReader
from llama_index.core import Document, PromptTemplate
from llama_index.llms.openai import OpenAI

from pycolbertdb.client import Colbertdb
from pycolbertdb.models import CreateCollectionDocument
from pycolbertdb.helpers import from_llama_index_documents

Load Environment Variables

Load your environment variables from a .env file.

load_dotenv()
URL = os.getenv('COLBERTDB_URL')
API_KEY = os.getenv('COLBERTDB_API_KEY')
STORE_NAME = os.getenv('COLBERTDB_STORE_NAME')
OPEN_AI_KEY = os.getenv('OPENAI_API_KEY')

URLS = ['https://en.wikipedia.org/wiki/Onigiri']

Initialize Clients

Initialize the ColbertDB and OpenAI clients.

client = Colbertdb(url=URL, api_key=API_KEY, store_name=STORE_NAME)
open_ai_client = OpenAI(model="gpt-4-turbo", api_key=OPEN_AI_KEY)

qa_prompt_tmpl_str = """\
Context information is below.
---------------------
{context_str}
---------------------
Given the context information and not prior knowledge, answer the query.
Please write the answer in the style of {tone_name}
Query: {query_str}
Answer: \
"""

prompt_tmpl = PromptTemplate(qa_prompt_tmpl_str)

Fetch and Process Documents

Fetch and process HTML content from the specified URLs.

docs = from_llama_index_documents(SimpleWebPageReader(html_to_text=True).load_data(URLS))

Create a Collection in ColbertDB

Create a new collection in ColbertDB with the processed documents.

collection = client.create_collection(documents=docs, name='rice_ball_facts', options={"force_create": True})

Search the Collection

Perform a search query on the created collection.

result = collection.search(query="What are some popular fillings for onigiri?", k=3)

Generate a Response Using OpenAI

Format the retrieved documents and generate a response using OpenAI.

context = ''
for document in result.documents:
    print("Source: " + document.metadata['source'] + "\n", document.content)
    context += (document.content + "\n\n")

prompt = prompt_tmpl.format(context_str=context, tone_name="shakespeare", query_str="What are some typical onigiri fillings")
response = open_ai_client.complete(prompt)
print(response)

Add New Documents to the Collection

Fetch additional documents and add them to the existing collection.

new_docs = SimpleWebPageReader(html_to_text=True).load_data(["https://en.wikipedia.org/wiki/Kewpie_(mayonnaise)"])
new_formatted = [{"content": doc.text, "metadata": {"source": doc.id_}} for doc in new_docs[0:2]]

collection = collection.add_documents(documents=new_formatted)

Search the Updated Collection

Perform a new search query on the updated collection.

new_result = collection.search(query="When was kewpie mayo founded?", k=3)
new_context = ''
for document in new_result.documents:
    print("Source: " + document.metadata['source'] + "\n", document.content)
    new_context += (document.content + "\n\n")

prompt = prompt_tmpl.format(context_str=new_context, tone_name="bruce springsteen", query_str="When and where was kewpie mayo founded")
new_response = open_ai_client.complete(prompt)
print(new_response)

Conclusion

This guide provides a quickstart overview of using the pycolbertdb package for document processing and querying. Customize the prompt and collection as needed for your specific use case.