Summary

IRIS-VectorLab is a web application that demonstrates the functionality of Vector Search with the help of embedded python. It leverages the functionality of the Python framework SentenceTransformers for state-of-the-art sentence embeddings and uses HuggingFace pipeline along with GPT2 LLM (Large Language Model) model for text generation.

Features

• Text to Embeddings Translation.
• VECTOR-typed Data Insertion.
• View Vector Data
• Perform Vector Search by using VECTOR_DOT_PRODUCT and VECTOR_COSINE functions.
• Demonstrate the difference between normal and vector search
• HuggingFace Text generation with the help of GPT2 LLM (Large Language Model) model and Hugging Face pipeline

Application Interface

Prerequisites

Make sure you have git and Docker desktop installed.

Installation

Docker (e.g. for dev purposes)

Clone/git pull the repo into any local directory

$ git clone https://github.com/mwaseem75/iris-VectorLab.git

Open the terminal in this directory and run:

$ docker-compose up -d

IPM

Open IRIS for Health installation with IPM client installed. Call in any namespace:

USER>zpm "install iris-VectorLab"

Application already loaded some dummy data for testing purposes.

Test functionality through IRIS terminal

Run the below command to start IRIS terminal from VS CODE terminal

$ docker-compose exec iris iris session iris

View existing data

Run the below command in IRIS terminal to view the existing data

do ##class(dc.VectorLab.Base).ListData()

Inserting vector data

Run the below command to insert vector data, Pass the desired description as an argument

do ##class(dc.VectorLab.Base).SaveData("InterSystems IRIS Data Platform 2024.1 introduces Vector Search")

Run the ListData command again to view the data

View vector data

Run the below command to view vector data in IRIS terminal and pass the ID which is 17 in our case.

set vector =  ##class(dc.VectorLab.Base).ViewData(17,2) 
write vector

Performing Vector Search

Run the below command to search vector data

set vector =  ##class(dc.VectorLab.Base).VectorSearch("The fox and the chicken")

Application will get the top 5 similar records

HuggingFace Text generation with the help of GPT2 LLM

Run the below command by passing some text, application will generate the text accordingly to search vector data

set rc=##class(dc.VectorLab.Base).Generate("Generative AI is")
write rc

Test the functionality by Running Web Application

Navigate to http://localhost:52773/csp/vectorlab/index.csp to run the application. (by using demo|demo)

Add data with vector embeddings

Click on the Resources List to view Resource records of the connected FHIR Server and further clink on the record itself to view JSON and Human readable details of selected Resource

View Vector Data

From application

From Management portal

Load data

If needed, Load data button will remove all the data and load the fresh data.

Performing Vector Search

Type “The fox and the chicken” in search and click Vector search button

Application will get the top 5 similar vector data.

Now we will do a traditional search by clicking the Normal search button, as expected system will not retrieve any record.

HuggingFace Text generation with the help of GPT2 LLM and Hugging Face pipeline

Navigate to http://localhost:52773/csp/vectorlab/generate.csp to run the application.

Type some text and press generate button, application will generate the text accordingly with the help of GPT2 LLM (Large Language Model) model and Hugging Face pipeline

Thanks