llamaindex-try

random attempts in learning llamaindex

Prerequisites

Python dependencies

1. (Optional) set up virtual environment. There are many ways. Here's an example

   $ python -m venv venv
   $ source venv/bin/activate

2. Install dependencies

   $ pip install -r requirements.txt

Install Qdrant database

1. Install Docker
2. Follow https://qdrant.tech/documentation/quick-start/ to install Qdrant

Environment variables

1. Set up your OpenAI API key

$ export OPENAI_API_KEY="sk-..."

2. Prepare your own dataset to be retrieved, and then set the environment variable pointing to the directory

$ export RAG_DATA_DIR="<path to data>"

My code assumes a bunch of .md files, but you can customize by messing around with SimpleDirectoryReader in the code

Execution

This repo is not a package. This repo contains separated codes, each of which representing a use case or a scenario.

To run it, simply run in the project directory and

$ PYTHONPATH=$(pwd) python src/<code.py>

Run Qdrant

Some codes require Qdrant. After you install Qdrant, you can use ./run-qdrant.sh to start it.

Run "rag_bot_1"

To run src/rag_bot_1, which runs a RAG Chat bot using streamlit, simple run ./run-rag_bot_1.sh

Try RAG

• rag_1.py - use the start example with my own documents in markdown
• rag_2.py - Customize LLM and prompts
• rag_3.py - Use a vector database (Qdrant) + Data ingestion
• rag_4.py - Create the query engine in a slightly lower level way
• rag_5.py - Instrumentation and customized event handler
• rag_6.py - Use an embedding model on HuggingFace
• rag_7.py - Use different chunking/indexing strategies
• rag_8.py - Use IngestionPipeline to parse doc and feed nodes into vector db
• rag_bot_1 - The most simple RAG chat app using Streamlit as UI