zhtet/document-chat

1# Reference https://huggingface.co/spaces/johnmuchiri/anspro1/blob/main/app.py2# Resource https://python.langchain.com/docs/modules/chains3 4import streamlit as st5from langchain_community.document_loaders.pdf import PyPDFLoader6from langchain.text_splitter import RecursiveCharacterTextSplitter7from langchain_community.vectorstores.pinecone import Pinecone8from langchain_openai import OpenAIEmbeddings, ChatOpenAI9from langchain.memory import ConversationBufferMemory10from langchain_core.prompts import ChatPromptTemplate11from langchain.chains import ConversationalRetrievalChain, RetrievalQAWithSourcesChain12import openai13from dotenv import load_dotenv14import os15 16import pinecone17 18load_dotenv()19 20# please create a streamlit app on huggingface that uses openai api21# and langchain data framework, the user should be able to upload22# a document and ask questions about the document, the app should23# respond with an answer and also display where the response is24# referenced from using some sort of visual annotation on the document25 26# set the path where you want to save the uploaded PDF file27SAVE_DIR = "pdf"28 29 30def generate_response(pages, query_text, k, chain_type):31    if pages:32        pinecone.init(33            api_key=os.getenv("PINECONE_API_KEY"),34            environment=os.getenv("PINECONE_ENV_NAME"),35        )36 37        vector_db = Pinecone.from_documents(38            documents=pages, embedding=OpenAIEmbeddings(), index_name="document-chat"39        )40 41        retriever = vector_db.as_retriever(42            search_type="similarity", search_kwards={"k": k}43        )44 45        prompt_template = ChatPromptTemplate.from_messages(46            [47                (48                    "system",49                    "You are a helpful assistant that can answer questions regarding to a document provided by the user.",50                ),51                ("human", "Hello, how are you doing?"),52                ("ai", "I'm doing well, thanks!"),53                ("human", "{user_input}"),54            ]55        )56 57        llm = ChatOpenAI(model_name="gpt-3.5-turbo", temperature=0)58 59        # create a chain to answer questions60        qa = RetrievalQAWithSourcesChain.from_chain_type(61            llm=llm,62            chain_type=chain_type,63            retriever=retriever,64            return_source_documents=True,65            # prompt_template=prompt_template,66        )67 68        response = qa({"question": query_text})69        return response70 71 72def visual_annotate(document, answer):73    # Implement this function according to your specific requirements74    # Highlight the part of the document where the answer was found75    start = document.find(answer)76    annotated_document = (77        document[:start]78        + "**"79        + document[start : start + len(answer)]80        + "**"81        + document[start + len(answer) :]82    )83    return annotated_document84 85 86st.set_page_config(page_title="🦜🔗 Ask the Doc App")87st.title("Document Question Answering App")88 89with st.sidebar.form(key="sidebar-form"):90    st.header("Configurations")91 92    openai_api_key = st.text_input("Enter OpenAI API key here", type="password")93    os.environ["OPENAI_API_KEY"] = openai_api_key94 95    pinecone_api_key = st.text_input(96        "Enter your Pinecone environment key", type="password"97    )98    os.environ["PINECONE_API_KEY"] = pinecone_api_key99 100    pinecone_env_name = st.text_input("Enter your Pinecone environment name")101    os.environ["PINECONE_ENV_NAME"] = pinecone_env_name102 103    submitted = st.form_submit_button(104        label="Submit",105        # disabled=not (openai_api_key and pinecone_api_key and pinecone_env_name),106    )107 108left_column, right_column = st.columns(2)109 110with left_column:111    uploaded_file = st.file_uploader("Choose a pdf file", type="pdf")112    pages = []113 114    if uploaded_file is not None:115        # save the uploaded file to the specified directory116        file_path = os.path.join(SAVE_DIR, uploaded_file.name)117        with open(file_path, "wb") as f:118            f.write(uploaded_file.getbuffer())119        st.success(f"File {uploaded_file.name} is saved at path {file_path}")120 121        loader = PyPDFLoader(file_path=file_path)122        text_splitter = RecursiveCharacterTextSplitter(chunk_size=500, chunk_overlap=0)123        pages = loader.load_and_split(text_splitter=text_splitter)124 125    query_text = st.text_input(126        "Enter your question:", placeholder="Please provide a short summary."127    )128 129    chain_type = st.selectbox(130        "chain type", ("stuff", "map_reduce", "refine", "map_rerank")131    )132 133    k = st.slider("Number of relevant chunks", 1, 5)134 135    with st.spinner("Retrieving and generating a response ..."):136        response = generate_response(137            pages=pages, query_text=query_text, k=k, chain_type=chain_type138        )139 140        with right_column:141            st.write("Output of your question")142 143            if response:144                st.subheader("Result")145                st.write(response["answer"])146                print("response: ", response)147 148                st.subheader("source_documents")149                for each in response["source_documents"]:150                    st.write("page: ", each.metadata["page"])151                    st.write("source: ", each.metadata["source"])152            else:153                st.write("response not showing at the moment")154 155 156# with st.form("myform", clear_on_submit=True):157#     openai_api_key = st.text_input(158#         "OpenAI API Key", type="password", disabled=not (uploaded_file and query_text)159#     )160#     submitted = st.form_submit_button(161#         "Submit", disabled=not (pages and query_text)162#     )163#     if submitted and openai_api_key.startswith("sk-"):164#         with st.spinner("Calculating..."):165#             response = generate_response(pages, openai_api_key, query_text)166#             result.append(response)167#             del openai_api_key168 169# if len(result):170#     st.info(response)171 172# if st.button("Get Answer"):173#     answer = get_answer(question, document)174#     st.write(answer["answer"])175 176#     # Visual annotation on the document177#     annotated_document = visual_annotate(document, answer["answer"])178#     st.markdown(annotated_document)179