RAG - Chroma multi-modal multi-vector

Multi-modal LLMs enable visual assistants that can perform question-answering about images.

This template create a visual assistant for slide decks, which often contain visuals such as graphs or figures.

It uses GPT-4V to create image summaries for each slide, embeds the summaries, and stores them in Chroma.

Given a question, relevant slides are retrieved and passed to GPT-4V for answer synthesis.

"Multi-modal LLM Process Diagram"

Input

Supply a slide deck as pdf in the /docs directory.

By default, this template has a slide deck about Q3 earnings from DataDog, a public technology company.

Example questions to ask can be:

How many customers does Datadog have?
What is Datadog platform % Y/Y growth in FY20, FY21, and FY22?

To create an index of the slide deck, run:

poetry install
python ingest.py

Storage

Here is the process the template will use to create an index of the slides (see blog):

• Extract the slides as a collection of images
• Use GPT-4V to summarize each image
• Embed the image summaries using text embeddings with a link to the original images
• Retrieve relevant image based on similarity between the image summary and the user input question
• Pass those images to GPT-4V for answer synthesis

By default, this will use LocalFileStore to store images and Chroma to store summaries.

For production, it may be desirable to use a remote option such as Redis.

You can set the local_file_store flag in chain.py and ingest.py to switch between the two options.

For Redis, the template will use UpstashRedisByteStore.

We will use Upstash to store the images, which offers Redis with a REST API.

Simply login here and create a database.

This will give you a REST API with:

• UPSTASH_URL
• UPSTASH_TOKEN

Set UPSTASH_URL and UPSTASH_TOKEN as environment variables to access your database.

We will use Chroma to store and index the image summaries, which will be created locally in the template directory.

LLM

The app will retrieve images based on similarity between the text input and the image summary, and pass the images to GPT-4V.

Environment Setup

Set the OPENAI_API_KEY environment variable to access the OpenAI GPT-4V.

Set UPSTASH_URL and UPSTASH_TOKEN as environment variables to access your database if you use UpstashRedisByteStore.

Usage

To use this package, you should first have the LangChain CLI installed:

pip install -U langchain-cli

To create a new LangChain project and install this as the only package, you can do:

langchain app new my-app --package rag-chroma-multi-modal-multi-vector

If you want to add this to an existing project, you can just run:

langchain app add rag-chroma-multi-modal-multi-vector

And add the following code to your server.py file:

from rag_chroma_multi_modal_multi_vector import chain as rag_chroma_multi_modal_chain_mv

add_routes(app, rag_chroma_multi_modal_chain_mv, path="/rag-chroma-multi-modal-multi-vector")

(Optional) Let's now configure LangSmith. LangSmith will help us trace, monitor and debug LangChain applications. You can sign up for LangSmith here. If you don't have access, you can skip this section

export LANGCHAIN_TRACING_V2=true
export LANGCHAIN_API_KEY=<your-api-key>
export LANGCHAIN_PROJECT=<your-project>  # if not specified, defaults to "default"

If you are inside this directory, then you can spin up a LangServe instance directly by:

langchain serve

This will start the FastAPI app with a server is running locally at http://localhost:8000

We can see all templates at http://127.0.0.1:8000/docs We can access the playground at http://127.0.0.1:8000/rag-chroma-multi-modal-multi-vector/playground

We can access the template from code with:

from langserve.client import RemoteRunnable

runnable = RemoteRunnable("http://localhost:8000/rag-chroma-multi-modal-multi-vector")