main.py

import os
import time
from operator import itemgetter
from collections import Counter
from langchain.schema.runnable import Runnable, RunnablePassthrough, RunnableLambda
from langchain.schema.runnable.config import RunnableConfig
from langchain.embeddings import HuggingFaceEmbeddings
from langchain.chains import ConversationalRetrievalChain
from langchain.callbacks.streaming_stdout import StreamingStdOutCallbackHandler
from langchain.prompts import ChatPromptTemplate, MessagesPlaceholder
from langchain.schema import StrOutputParser
from langchain.chains.qa_with_sources import load_qa_with_sources_chain
from langchain_pinecone import PineconeVectorStore

from pinecone import Pinecone
from langchain.memory import ChatMessageHistory, ConversationBufferMemory
import pandas as pd
import numpy as np
from langchain_core.messages import HumanMessage, SystemMessage
from langchain_core.prompts.chat import (
    ChatPromptTemplate,
    HumanMessagePromptTemplate,
    SystemMessagePromptTemplate,
)
from langchain_openai import ChatOpenAI
import chainlit as cl
from chainlit.input_widget import TextInput
from chainlit import user_session
from offres_emploi import Api
from offres_emploi.utils import dt_to_str_iso
import datetime
import bcrypt
import json

@cl.password_auth_callback
def auth_callback(username: str, password: str):
    auth = json.loads(os.environ['CHAINLIT_AUTH_LOGIN'])
    ident = next(d['ident'] for d in auth if d['ident'] == username)
    pwd = next(d['pwd'] for d in auth if d['ident'] == username)
    resultLogAdmin = bcrypt.checkpw(username.encode('utf-8'), bcrypt.hashpw(ident.encode('utf-8'), bcrypt.gensalt())) 
    resultPwdAdmin = bcrypt.checkpw(password.encode('utf-8'), bcrypt.hashpw(pwd.encode('utf-8'), bcrypt.gensalt())) 
    resultRole = next(d['role'] for d in auth if d['ident'] == username)
    if resultLogAdmin and resultPwdAdmin and resultRole == "admindatapcc":
        return cl.User(
            identifier=ident + " : 🧑‍💼 Admin Datapcc", metadata={"role": "admin", "provider": "credentials"}
        )
    elif resultLogAdmin and resultPwdAdmin and resultRole == "userdatapcc":
        return cl.User(
            identifier=ident + " : 🧑‍🎓 User Datapcc", metadata={"role": "user", "provider": "credentials"}
        )
        
os.environ["TOKENIZERS_PARALLELISM"] = os.environ["TOKENIZERS_PARALLELISM"]
os.environ['OPENAI_API_KEY'] = os.environ['OPENAI_API_KEY']

@cl.author_rename
def rename(orig_author: str):
    rename_dict = {"DatapccSkillStream": "Datapcc", "ConversationalRetrievalChain": "Assistant conversationnel 💬", "Retriever": "Agent conversationnel", "StuffDocumentsChain": "Chaîne de documents", "LLMChain": "Agent", "ChatOpenAI": "IA🤖"}
    return rename_dict.get(orig_author, orig_author)

@cl.action_callback("download")
async def on_action(action):
    content = []
    content.append(action.value)
    arrayContent = np.array(content)
    df = pd.DataFrame(arrayContent)
    with open('./' + action.description + '.txt', 'wb') as csv_file:
        df.to_csv(path_or_buf=csv_file, index=False,header=False, encoding='utf-8')
    elements = [
        cl.File(
            name= action.description + ".txt",
            path="./" + action.description + ".txt",
            display="inline",
        ),
    ]
    await cl.Message(
        author="Datapcc 🌐🌐🌐", content="[Lien] 🔗", elements=elements
    ).send()
    await action.remove()

def retriever_to_cache():
    os.environ['PINECONE_API_KEY'] = os.environ['PINECONE_API_KEY']
    os.environ['PINECONE_ENVIRONMENT'] = "us-west4-gcp-free"
    index_name = os.environ['PINECONE_INDEX_NAME']
    embeddings = HuggingFaceEmbeddings()
    vectorstore = PineconeVectorStore(
        index_name=index_name, embedding=embeddings
    )
    retriever = vectorstore.as_retriever(search_type="similarity_score_threshold", search_kwargs={"score_threshold": .7, "k": 30,"filter": {'categorie': {'$eq': 'OF'}}})
    return retriever

@cl.set_chat_profiles
async def chat_profile():
    return [
        cl.ChatProfile(name="OF - Offre de formation",markdown_description="Requêter sur l'offre de formation - OF",icon="./public/favicon.png",),
    ]
@cl.on_chat_start
async def start():
    chat_profile = cl.user_session.get("chat_profile")
    chatProfile = chat_profile.split(' - ')
    
    if chatProfile[0] == 'OF':
        app_user = cl.user_session.get("user")
        welcomeUser = app_user.identifier
        welcomeUserArray = welcomeUser.split('@')
        welcomeUserStr = welcomeUserArray[0].replace('.',' ')
        await cl.Message(f"> Bonjour {welcomeUserStr}").send()
        await cl.Message(
            author="Datapcc 🌐🌐🌐",content=f"✨ Commencez à poser vos questions sur les données \"{chat_profile}\"\n- Création de BCC à partir d'une liste de savoirs ou d'objectifs pédagogiques\n- Création du tableau de la version n°1 de la maquette de formation"
        ).send()
    
        from langchain_core.prompts.prompt import PromptTemplate

        _template = """Given the following conversation and a follow up question, rephrase the follow up question to be a standalone question, in its original language.

                    Chat History:
                    {chat_history}
                    Follow Up Input: {question}
                    Standalone question:"""
        CONDENSE_QUESTION_PROMPT = PromptTemplate.from_template(_template)
        
        ########## Chain with streaming ##########
        message_history = ChatMessageHistory()
        memory = ConversationBufferMemory(
            memory_key="chat_history",
            output_key="answer",
            chat_memory=message_history,
            return_messages=True,
        )
        streaming_llm = ChatOpenAI(
            model_name = "gpt-4-1106-preview",
            streaming=True,
            temperature=1
        )
        qa = ConversationalRetrievalChain.from_llm(
            streaming_llm,
            memory=memory,
            chain_type="stuff",
            return_source_documents=True,
            verbose=False,
            retriever=retriever_to_cache()
        )
        cl.user_session.set("conversation_chain", qa)
    
@cl.on_message
async def main(message: cl.Message):
    chat_profile = cl.user_session.get("chat_profile")
    chatProfile = chat_profile.split(' - ')
    if chatProfile[0] == "OF":
        chain = cl.user_session.get("conversation_chain")
        cb = cl.AsyncLangchainCallbackHandler()
        res = await chain.acall("Contexte : Réponds à la question suivante de la manière la plus pertinente, la plus exhaustive et la plus détaillée possible, dans le contexte et les informations fournies. Question : " + message.content, callbacks=[cb])
        answer = res["answer"]
        source_documents = res["source_documents"]
    
        text_elements = []
        metadatas = ''
        if source_documents:
            for source_idx, source_doc in enumerate(source_documents):
                numSource = source_idx + 1
                source_name = f"Source n°{numSource}"
                text_elements.append(
                    cl.Text(content="Formations : " + source_doc.metadata['ABREGE_LIBELLES'] + " " + source_doc.metadata['INTITULE'] + "\n\nROME : " + source_doc.metadata['CODES_ROME'] + "\nLibellés ROME : " + source_doc.metadata['LIBELLES_ROME'] + "\n\nActivités : " + source_doc.metadata['ACTIVITES_VISEES'].replace('œ','oe') + "\n\nEmplois accessibles : " + source_doc.metadata['TYPE_EMPLOI_ACCESSIBLES'] + "\n\nCompétences : " + source_doc.metadata['CAPACITES_ATTESTEES'].replace('œ','oe').replace('…','oe'), name=source_name)
                )
            source_names = [text_el.name for countMetadata, text_el in enumerate(text_elements) if countMetadata < 10]
            if source_names:
                metadatas += ', '.join(source_names)
            else:
                metadatas += "\n\nPas de source trouvée!"

        actions = [
            cl.Action(name="download", value="Question : " + message.content + "\n\nRéponse : " + answer, description="download_offre_formation")
        ]

        await cl.Message(author="Datapcc 🌐🌐🌐",content=answer).send()
        await cl.Message(author="Datapcc 🌐🌐🌐",content="Download", actions=actions).send()
    
        if metadatas:
            await cl.Message(author="Datapcc 🌐🌐🌐",content="Sources : " + metadatas, elements=text_elements).send()