app.py

import os

import gradio as gr
import openai
from pathlib import Path

from langchain import PromptTemplate, LLMChain
import qdrant_client
from dotenv import load_dotenv
# from langchain.chains.qa_with_sources import load_qa_with_sources_chain
from langchain.chains import RetrievalQA
from langchain.chat_models import ChatOpenAI
from langchain.chains.question_answering import load_qa_chain
from langchain.embeddings.openai import OpenAIEmbeddings
from langchain.llms import AzureOpenAI, OpenAI
from langchain.text_splitter import CharacterTextSplitter
from langchain.vectorstores import Qdrant
from qdrant_client import QdrantClient

# Constants
collection_name="10ks"
# collection_name="collectiveagreements"


# Load the environment variables with the Azure OpenAI API key
load_dotenv()

# Initialize Azure OpenAI
# openai.api_type = os.getenv("OPENAI_API_TYPE")
# openai.api_base = os.getenv("OPENAI_API_BASE")
openai.api_key = os.getenv("OPENAI_API_KEY")
# openai.api_version = os.getenv("OPENAI_API_VERSION")

# The data was vectorized with ADA, so we'll use that to convert our
# query into a vector
embeddings = OpenAIEmbeddings(model="text-embedding-ada-002") #, chunk_size=1)

dbclient = QdrantClient("localhost", port=6333, grpc_port=6334, prefer_grpc=True)

index=Qdrant(client=dbclient, collection_name=collection_name, embeddings=embeddings, vector_name="fragmentvector")

# # Load the FAISS index
# index = dbclient.retrieve. .load_local(
#     "collectiveagreements.db",
#     OpenAIEmbeddings(chunk_size=1, model="text-embedding-ada-002"),
# )

# Open a connection to render the search results into test- this uses davinci-002.
llm = OpenAI(deployment_name="davinci", temperature=0)

# Open op a connection to do the querying
# Chain type can be stuff, map_reduce or refine
# chain = load_qa_with_sources_chain(llm, chain_type="map_reduce")

def docquery(question):
    docs = index.similarity_search(question)
    print("Length of answer: ", len(docs))
    # Process the query and return the results
    llm = ChatOpenAI(model_name="gpt-3.5-turbo", temperature=0)
    qa_chain = RetrievalQA.from_chain_type(llm,retriever=index.as_retriever(), return_source_documents=True)
    output=qa_chain({"query": question})
    # output = chain.run(input_documents=docs, question=question)
    print("Is the error here?", output  )
    # myanswer = "##" + output.split("SOURCES")[0]
    references = ""
    # print("Docs:", docs)
    for i in docs:
        print("item: ", i.page_content)
        references = (
            references
            + "**"
            + "** \n"
            + i.page_content.replace("\n", "")
            + "\n\n"
        )
    return output['result'], references

with gr.Blocks(title="Collective Agreement Search") as blocks:
    appname = gr.Markdown(value="# 10K filings search")
    appdesc = gr.Markdown(
        value="## The tabs below demonstration different ways to query the data."
    )

    with gr.Tab("Ask a question"):
        appdesc = gr.Markdown(
            value="### This is a demo of an OpenAI-based question answering system. Type in a question and the system will return the answer and the source document."
        )
        question = gr.Textbox(
            lines=1,
            label="Question: press enter to submit",
            value="Where is Babcock's head office?",
        )
        answer = gr.Markdown(label="Answer")
        references = gr.Markdown(label="References")
        question.submit(docquery, question, outputs=[answer, references])


blocks.launch(share=True, server_name="0.0.0.0", server_port=8080)