README

Table of Contents

Overview

This repo contains a sample RAG based solution in promptflow using the SDK.

Prerequisites

• Python 3.9 or Python 3.10
• promptflow sdk
• OpenAI API Key or Azure OpenAI APIP Key
• Azure Cognitive Search

Getting Started

Setup

To install the dependencies, run the following command:

pip install -r requirements.txt

You wll also need to create a .env file. To do so, copy the .env.sample file and rename it to .env. Then, fill in the values for the following variables:

• AZURE_SEARCH_ENDPOINT [Required]: The Azure Cognitive Search Endpoint
• AZURE_SEARCH_ADMIN_KEY [Required]: The Azure Cognitive Search Admin Key
• INDEX_NAME [Required]: The name of the index in Azure Cognitive Search
• OPENAI_API_TYPE [Required]: The OpenAI Type (either azure or openai)
• OPENAI_API_BASE [Optional]: The OpenAI base url
• OPENAI_API_KEY [Required]: The OpenAI API Key
• OPENAI_API_VERSION [Default=2023-05-15]: The version of OpenAI API

Creating the Connections

Connection are ways for promptflow to interact with services in a secure way. Since this sample interfaces with Azure Cognitive Search and OpenAI, we need to establish safe and secure ways to use the services.

To create the Azure Cognitive Search connection run the following:

pf connection create --file src/connections/cognitive_search.yaml --set api_key="<your-api-key>" --set api_base="<your-search-endpoint>"

To create the Azure OpenAI connection run the following:

pf connection create --file src/connections/azure_openai.yaml --set api_key="<your-api-key>" --set api_base="<your-aoai-endpoint>"

NOTE: The process to create a OpenAI connection is very similar, but you will need to create a openai connection and reference it in the flow.dag.yaml file.

Creating your Index

To create the index and populate the data in cognitive search, run the script with the following command:

python -m src.scripts.chunk_and_upload

This script will both create the index (based on the index file name in your .env file) and will upload the chunked data from the langchain introduction file.

Testing the Flow

To test the flow, run the following command:

pf flow test --flow src/flows/standard/promptflow-rag --inputs text="What is Langchain?"

Running the Flow

To run the flow, run the following command:

pf run create --flow --flow src/flows/standard/promptflow-rag --data src/flows/standard/promptflow-rag/data.jsonl