app.py

from langchain.embeddings.openai import OpenAIEmbeddings
from langchain.text_splitter import RecursiveCharacterTextSplitter, CharacterTextSplitter
from langchain.vectorstores import Chroma
from langchain.chains import RetrievalQAWithSourcesChain
from langchain.memory import ConversationBufferWindowMemory
from langchain.chains import ConversationalRetrievalChain
from langchain.chat_models import ChatOpenAI
from langchain.prompts.chat import (
    ChatPromptTemplate,
    SystemMessagePromptTemplate,
    HumanMessagePromptTemplate,
)
from langchain.document_loaders import PyPDFLoader
import os
import chainlit as cl
from langchain.prompts import PromptTemplate

text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=100)

system_template = """Use the following pieces of context to answer the users question.
If you don't know the answer, just say that you don't know, don't try to make up an answer.
ALWAYS return a "SOURCES" part in your answer.
The "SOURCES" part should be a reference to the source of the document from which you got your answer.

Example of your response should be:

```
The answer is foo
SOURCES: xyz
```

Begin!
----------------
{summaries}"""
messages = [
    SystemMessagePromptTemplate.from_template(system_template),
    HumanMessagePromptTemplate.from_template("{question}"),
]
prompt = ChatPromptTemplate.from_messages(messages)
chain_type_kwargs = {"prompt": prompt}

@cl.on_chat_start
async def start():
    await cl.Avatar(
        name="ChatPDF",
        url="https://avatars.githubusercontent.com/u/128686189?s=400&u=a1d1553023f8ea0921fba0debbe92a8c5f840dd9&v=4",
        # path = r'assets/ChatPDFAvatar.jpg'
    ).send()


@cl.langchain_factory(use_async=True)
async def init():
    files = None

    # Wait for the user to upload a file
    while files == None:
        files = await cl.AskFileMessage(
            content="Hey, Welcome to ChatPDF!\n\nChatPDF is a smart, user-friendly tool that integrates state-of-the-art AI models with text extraction and embedding capabilities to create a unique, conversational interaction with your PDF documents.\n\nSimply upload your PDF, ask your questions, and ChatPDF will deliver the most relevant answers directly from your document.\n\nPlease upload a PDF file to begin!",max_size_mb=100, accept=["application/pdf"]
        ).send()

    file = files[0]

    msg = cl.Message(content=f'''Processing "{file.name}"...''')
    await msg.send()

    #
    
    with open(os.path.join(file.name), "wb") as f:
        f.write(file.content)

    print(file.name)
    
    loader = PyPDFLoader(file.name)
    pages = loader.load_and_split()

    # add page split info
    # Initialize a dictionary to keep track of duplicate page numbers
    page_counts = {}

    for document in pages:
        page_number = document.metadata['page']

        # If this is the first occurrence of this page number, initialize its count to 1
        # Otherwise, increment the count for this page number
        page_counts[page_number] = page_counts.get(page_number, 0) + 1

        # Create the page split info string
        page_split_info = f"Page-{page_number+1}.{page_counts[page_number]}"

        # Add the page split info to the document's metadata
        document.metadata['page_split_info'] = page_split_info

    

    # Create a Chroma vector store
    embeddings = OpenAIEmbeddings()
    docsearch = await cl.make_async(Chroma.from_documents)(
        pages, embeddings
    )

    # define memory
    memory = ConversationBufferWindowMemory(
        k=5,
        memory_key='chat_history', 
        return_messages=True, 
        output_key='answer'
        )
    
    # Create a chain that uses the Chroma vector store
    chain = ConversationalRetrievalChain.from_llm(
        ChatOpenAI(temperature=0, model="gpt-4", streaming=True),
        chain_type="stuff",
        retriever=docsearch.as_retriever(search_kwargs={'k':5}),
        memory=memory,
        return_source_documents=True,
    )

    # Save the metadata and texts in the user session
    # cl.user_session.set("metadatas", metadatas)
    cl.user_session.set("texts", pages)

    # Let the user know that the system is ready
    await msg.update(content=f''' "{file.name}" processed. You can now ask questions!''')


    return chain


@cl.langchain_postprocess
async def process_response(res):
    answer = res["answer"]
    source_documents = res['source_documents']
    content = [source_documents[i].page_content for i in range(len(source_documents))]
    name = [source_documents[i].metadata['page_split_info'] for i in range(len(source_documents))]
    source_elements = [
        cl.Text(content=content[i], name=name[i]) for i in range(len(source_documents))
        ]
    
    if source_documents:
        answer += f"\n\nSources: {', '.join([source_documents[i].metadata['page_split_info'] for i in range(len(source_documents))])}"
    else:
        answer += "\n\nNo sources found"
    
    await cl.Message(content=answer, elements=source_elements).send()
    # await cl.Message(content=answer).send()