*Disclaimer - Domino Reference Projects are starter kits built by Domino researchers. They are not officially supported by Domino. Once loaded, they are yours to use or modify as you see fit. We hope they will be a beneficial tool on your journey!

OpenAI custom Q&A Reference Project

This reference project shows how to use OpenAI's LLM to do Q&A over information that OpenAI's models have not been trained on and will not be able to provide answers out of the box. The way this works is to create embeddings of the document(s) that you want to query, run a semantic search to return information that can be provided as context/information along with the user's query as a prompt to the LLM and get results back. The project has the following files

• OpenAI_QA_Pinecone.ipynb : This file loads a PDF,converts it to embeddings, stores the embeddings in Pinecone, runs the semantic search against the embeddings, constructs a prompt and calls OpenAI's models to get a response. You will need your OpenAPI and Pinecone keys to be set in the environment for this example. To work with OpenAI, set up your Pinecone index to have 1536 dimensions.

• OpenAI_QA_FAISS.ipynb : This file loads a PDF, converts it to embeddings, stores the embeddings locally using a FAISS index, runs the semantic search against the embeddings, constructs a prompt and calls OpenAI's models to get a response. You will need your OpenAPI key to be set in the environment for this example.

• OpenAI_QA_DDL_Docs.ipynb : This file loads a documents from a Domino Dataset, converts it to embeddings, stores the embeddings locally using a FAISS index (there is code you can uncomment if you want to use Pinecone), runs the semantic search against the embeddings, constructs a prompt and calls OpenAI's models to get a response. You will need your OpenAPI key to be set in the environment for this example.

• faiss_ddl_doc_store.pkl : This file contains the FAISS embeddings of Domino's documentation . You can use this if you don't want to (re)compute embeddings of Select_Global_Value_Fund.pdf again

• app.sh : The shell script needed to run the chat app

• app.py : Streamlit app code for the Q&A chatbot. This app uses faiss_ddl_doc_store.pkl for the embeddings

• Select_Global_Value_Fund.pdf : A report that can be used as an example for the flow that has been described above in case you want to compute embeddings on a fresh document

• Solution_Overview.pdf : A diagram that depicts the different components and the flow of information between them

Setup instructions

This project requires the following compute environments to be present:

PromptEngineering

Environment Base

Domino Standard Environment Py3.9 R4.2

Dockerfile Instructions

USER root

RUN sudo apt -y install tesseract-ocr
RUN sudo apt-get -y install libpoppler-dev
RUN sudo apt-get install poppler-utils
RUN sudo apt -y install libmagic-dev

RUN pip install pinecone-client
RUN pip install langchain==0.0.144
RUN pip install unstructured[local-inference]
RUN pip install poppler-utils
RUN pip install openai
RUN pip install faiss-cpu

RUN pip install streamlit && \
    pip install streamlit-chat && \
    pip install tiktoken 


RUN pip install "detectron2@git+https://github.com/facebookresearch/detectron2.git@v0.6#egg=detectron2"


USER ubuntu