app.py

from langchain.prompts.prompt import PromptTemplate
from langchain.llms import OpenAIChat
from langchain.chains import ConversationalRetrievalChain, LLMChain
from langchain.chains.question_answering import load_qa_chain
from langchain.embeddings import HuggingFaceEmbeddings, HuggingFaceInstructEmbeddings
from langchain.callbacks import StdOutCallbackHandler
from langchain.chains.conversational_retrieval.prompts import CONDENSE_QUESTION_PROMPT
from langchain.vectorstores import FAISS
from langchain.memory import ConversationBufferMemory
import os
from typing import Optional, Tuple
import gradio as gr
import pickle
from threading import Lock

from langchain.prompts.chat import (
    ChatPromptTemplate,
    SystemMessagePromptTemplate,
    AIMessagePromptTemplate,
    HumanMessagePromptTemplate,
)
from langchain.schema import (
    AIMessage,
    HumanMessage,
    SystemMessage
)

from langchain.prompts import PromptTemplate

prefix_messages = [{"role": "system", "content": "You are a helpful assistant that is very good at answering questions about investments using the information given."}]

model_options = {'all-mpnet-base-v2': "sentence-transformers/all-mpnet-base-v2",
                'instructor-base': "hkunlp/instructor-base"}

model_options_list = list(model_options.keys())

memory = ConversationBufferMemory(memory_key='chat_history', return_messages=True, output_key='answer')

def load_prompt():

    system_template="""Use only the following pieces of context that has been scraped from a website to answer the users question accurately.
    Do not use any information not provided in the website context.
    If you don't know the answer, just say 'There is no relevant answer in the Investor Website', 
    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.
    
    Remember, do not reference any information not given in the context.
    If the answer is not available in the given context just say 'There is no relevant answer in the website content'
    
    Follow the below format when answering:
    
    Question: {question}
    SOURCES: [xyz]
      
    Begin!
    ----------------
    {context}"""
    
    messages = [
        SystemMessagePromptTemplate.from_template(system_template),
        HumanMessagePromptTemplate.from_template("{question}")
    ]
    prompt = ChatPromptTemplate.from_messages(messages)
        
    return prompt
    
def load_vectorstore(model):
    '''load embeddings and vectorstore'''

    if 'mpnet' in model:
        
        emb = HuggingFaceEmbeddings(model_name=model)
        return FAISS.load_local('vanguard-embeddings', emb)

    elif 'instructor'in model:
        
        emb = HuggingFaceInstructEmbeddings(model_name=model,
                                               query_instruction='Represent the Financial question for retrieving supporting paragraphs: ',
                                               embed_instruction='Represent the Financial paragraph for retrieval: ')
        return FAISS.load_local('vanguard_embeddings_inst', emb)

#default embeddings and store
vectorstore = load_vectorstore(model_options['all-mpnet-base-v2'])

def on_value_change(change):
    '''When radio changes, change the embeddings'''
    global vectorstore
    vectorstore = load_vectorstore(model_options[change])
    
# vectorstore = load_vectorstore('vanguard-embeddings',sbert_emb)
    
_template = """Given the following conversation and a follow up question, rephrase the follow up question to be a standalone question.
You can assume the question about investing and the investment management industry.
Chat History:
{chat_history}
Follow Up Input: {question}
Standalone question:"""
CONDENSE_QUESTION_PROMPT = PromptTemplate.from_template(_template)

template = """You are an AI assistant for answering questions about investing and the investment management industry.
You are given the following extracted parts of a long document and a question. Provide a conversational answer.
If you don't know the answer, just say "Hmm, I'm not sure." Don't try to make up an answer.
If the question is not about investing, politely inform them that you are tuned to only answer questions about investing and the investment management industry.
Question: {question}
=========
{context}
=========
Answer in Markdown:"""
QA_PROMPT = PromptTemplate(template=template, input_variables=["question", "context"])


def get_chain(vectorstore):
    llm = OpenAIChat(streaming=True, 
                     callbacks=[StdOutCallbackHandler()], 
                     verbose=True, 
                     temperature=0,
                     model_name='gpt-4-0613')

    question_generator = LLMChain(llm=llm, prompt=CONDENSE_QUESTION_PROMPT)
    doc_chain = load_qa_chain(llm=llm,chain_type="stuff",prompt=load_prompt())

    chain = ConversationalRetrievalChain(retriever=vectorstore.as_retriever(search_kwags={"k": 3}), 
                                     question_generator=question_generator, 
                                     combine_docs_chain=doc_chain, 
                                     memory=memory, 
                                     return_source_documents=True, 
                                     get_chat_history=lambda h :h)

    
    return chain

def load_chain():
    chain = get_chain(vectorstore)
    return chain
    
    
class ChatWrapper:

    def __init__(self):
        self.lock = Lock()
    def __call__(
        self, inp: str, history: Optional[Tuple[str, str]], chain
    ):
        """Execute the chat functionality."""
        self.lock.acquire()
        try:
            history = history or []
            # Set OpenAI key
            # chain = get_chain(vectorstore)
            # Run chain and append input.
            output = chain({"question": inp})["answer"]
            history.append((inp, output))
        except Exception as e:
            raise e
        finally:
            self.lock.release()
        return history, history

block = gr.Blocks(css=".gradio-container {background-color: lightgray}")

with block:
    with gr.Row():
        gr.Markdown("<h3><center>Chat-Your-Data (Investor Education)</center></h3>")
        embed_but = gr.Button(value='Load QA Chain')

    with gr.Row():
        embeddings = gr.Radio(choices=model_options_list,value=model_options_list[0], label='Choose your Embedding Model',
                             interactive=True)
        embeddings.change(on_value_change, embeddings)
    
    vectorstore = load_vectorstore(embeddings.value)
    
    chatbot = gr.Chatbot()

    chat = ChatWrapper()


    with gr.Row():
        message = gr.Textbox(
            label="What's your question?",
            placeholder="Ask questions about Investing",
            lines=1,
        )
        submit = gr.Button(value="Send", variant="secondary").style(full_width=False)

    gr.Examples(
        examples=[
            "What are the benefits of investing in ETFs?",
            "What is the average cost of investing in a managed fund?",
            "At what age can I start investing?",
            "Do you offer investment accounts for kids?"
        ],
        inputs=message,
    )

    gr.HTML("Demo application of a LangChain chain.")

    gr.HTML(
        "<center>Powered by <a href='https://github.com/hwchase17/langchain'>LangChain 🦜️🔗</a></center>"
    )

    state = gr.State()
    agent_state = gr.State()

    
    submit.click(chat, inputs=[message, state, agent_state], outputs=[chatbot, state])
    message.submit(chat, inputs=[message, state, agent_state], outputs=[chatbot, state])

    embed_but.click(
        load_chain,
        outputs=[agent_state],
    )

    gr.Markdown("![visitor badge](https://visitor-badge.glitch.me/badge?page_id=nickmuchi-investor-chatchain)")

block.launch(debug=True)