import time
print('1')
print(time.time())

#__import__('pysqlite3')
#import sys
#sys.modules['sqlite3'] = sys.modules.pop('pysqlite3')

import os
import torch



#os.system('wget -q https://github.com/PanQiWei/AutoGPTQ/releases/download/v0.4.2/auto_gptq-0.4.2+cu118-cp310-cp310-linux_x86_64.whl')
#os.system('pip install -qqq auto_gptq-0.4.2+cu118-cp310-cp310-linux_x86_64.whl --progress-bar off')

#print(f"Is CUDA available: {torch.cuda.is_available()}")
os.system('nvidia-smi')

import uuid
#import replicate
import requests
import streamlit as st
from streamlit.logger import get_logger
from auto_gptq import AutoGPTQForCausalLM
from langchain import HuggingFacePipeline, PromptTemplate
from langchain.chains import RetrievalQA
from langchain.document_loaders import PyPDFDirectoryLoader
from langchain.embeddings import HuggingFaceInstructEmbeddings
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.vectorstores import Chroma
from pdf2image import convert_from_path
from transformers import AutoTokenizer, TextStreamer, pipeline
from langchain.memory import ConversationBufferMemory
from gtts import gTTS
from io import BytesIO
from langchain.chains import ConversationalRetrievalChain
import streamlit.components.v1 as components
#from sentence_transformers import SentenceTransformer
from langchain.document_loaders import UnstructuredMarkdownLoader
from langchain.vectorstores.utils import filter_complex_metadata
import fitz
from PIL import Image
from langchain.vectorstores import FAISS
import transformers

user_session_id = uuid.uuid4()

logger = get_logger(__name__)
st.set_page_config(page_title="Document QA by Dono", page_icon="🤖",  )
st.session_state.disabled = False
st.title("Document QA by Dono")
#st.markdown(f"""<style>
#            .stApp {{background-image: url("https://media.istockphoto.com/id/450481545/photo/glowing-lightbulb-against-black-background.webp?b=1&s=170667a&w=0&k=20&c=fJ91chWN1UkoKTNUvwgiQwpM80DlRpVC-WlJH_78OvE=");
#                     background-attachment: fixed;
#                     background-size: cover}}
#         </style>
#         """, unsafe_allow_html=True)

DEVICE = "cuda:0" if torch.cuda.is_available() else "cpu"


@st.cache_data
def load_data():
    loader = PyPDFDirectoryLoader("/home/user/app/pdfs/")
    docs = loader.load()
    print(len(docs))
    return docs



@st.cache_resource
def load_model(_docs):
    #embeddings = HuggingFaceInstructEmbeddings(model_name="hkunlp/instructor-large",model_kwargs={"device":DEVICE})
    embeddings = HuggingFaceInstructEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2",model_kwargs={"device":DEVICE})
    print(DEVICE)

    text_splitter = RecursiveCharacterTextSplitter(chunk_size=1024, chunk_overlap=256)
    texts = text_splitter.split_documents(docs)

    print('embedding done')
    
    #db = Chroma.from_documents(texts, embeddings, persist_directory="/home/user/app/db")
    db = FAISS.from_documents(texts, embeddings)

    print('db done')
    
    
    model_name_or_path = "TheBloke/Llama-2-13B-chat-GPTQ"
    model_basename = "model"

    tokenizer = AutoTokenizer.from_pretrained(model_name_or_path, use_fast=True)

    model = AutoGPTQForCausalLM.from_quantized(
        model_name_or_path,
        revision="gptq-8bit-128g-actorder_False",
        model_basename=model_basename,
        use_safetensors=True,
        trust_remote_code=True,
        inject_fused_attention=False,
        device=DEVICE,
        quantize_config=None,
    )

    print('model done')
    
    DEFAULT_SYSTEM_PROMPT = """
    You are a helpful, respectful and honest assistant. Always answer as helpfully as possible, while being safe. 
    Your answers should not include any harmful, unethical, racist, sexist, toxic, dangerous, or illegal content. 
    Please ensure that your responses are socially unbiased and positive in nature. 
    Always provide the citation for the answer from the text. 
    Try to include any section or subsection present in the text responsible for the answer. 
    Provide reference. Provide page number, section, sub section etc from which answer is taken.
    If a question does not make any sense, or is not factually coherent, explain why instead of answering something not correct. If you don't know the answer to a question, please don't share false information. 
    Given a government document that outlines rules and regulations for a specific industry or sector, use your language model to answer questions about the rules and their applicability over time. 
    The document may include provisions that take effect at different times, such as immediately upon publication, after a grace period, or on a specific date in the future. 
    Your task is to identify the relevant rules and determine when they go into effect, taking into account any dependencies or exceptions that may apply.
    The current date is 14 September, 2023. Try to extract information which is closer to this date and not in very past.
    Take a deep breath and work on this problem step-by-step. 
    """.strip()


    def generate_prompt(prompt: str, system_prompt: str = DEFAULT_SYSTEM_PROMPT) -> str:
        return f"""[INST] <<SYS>>{system_prompt}<</SYS>>{prompt} [/INST]""".strip()

    streamer = TextStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)

    text_pipeline = pipeline("text-generation",model=model,tokenizer=tokenizer,max_new_tokens=1024,
        temperature=0.2,top_p=0.95,repetition_penalty=1.15,streamer=streamer,)

    llm = HuggingFacePipeline(pipeline=text_pipeline, model_kwargs={"temperature": 0.2})

    print('llm done')
    
    SYSTEM_PROMPT = "Use the following pieces of context to answer the question at the end. If you don't know the answer, just say that you don't know, don't try to make up an answer."

    template = generate_prompt("""{context}  Question: {question} """,system_prompt=SYSTEM_PROMPT,) #Enter memory here!

    prompt = PromptTemplate(template=template, input_variables=["context",  "question"]) #Add history here

    qa_chain = RetrievalQA.from_chain_type(
        llm=llm,
        chain_type="stuff",
        retriever=db.as_retriever(search_kwargs={"k": 5}),
        return_source_documents=True,
        chain_type_kwargs={"prompt": prompt,
                           "verbose": False,
                           #"memory": ConversationBufferMemory(
                              #memory_key="history",
                              #input_key="question",
                              #return_messages=True)
                              },)

    print('load done')
    return qa_chain


#uploaded_file = len(docs)
#flag = 0
#if uploaded_file is not None:
#    flag = 1 

model_name_or_path = "TheBloke/Llama-2-13B-chat-GPTQ"
model_basename = "model"

st.session_state["llm_model"] = model_name_or_path


if "messages" not in st.session_state:
    st.session_state.messages = []



for message in st.session_state.messages:
    with st.chat_message(message["role"]):
        st.markdown(message["content"])


def on_select():
    st.session_state.disabled = True


def get_message_history():
    for message in st.session_state.messages:
        role, content = message["role"], message["content"]
        yield f"{role.title()}: {content}"


docs = load_data()
qa_chain = load_model(docs)




print('2')
print(time.time())
if prompt := st.chat_input("How can I help you today?"):
    st.session_state.messages.append({"role": "user", "content": prompt})
    with st.chat_message("user"):
        st.markdown(prompt)
    with st.chat_message("assistant"):
        message_placeholder = st.empty()
        full_response = ""
        message_history = "\n".join(list(get_message_history())[-3:])
        logger.info(f"{user_session_id} Message History: {message_history}")
        # question = st.text_input("Ask your question", placeholder="Try to include context in your question",
        # disabled=not uploaded_file,)
        print('3')
        print(time.time())
        result = qa_chain(prompt)
        print('4')
        print(time.time())
        
        output = [result['result']]

    # for item in output:
    #     full_response += item
    #     message_placeholder.markdown(full_response + "▌")
    #     message_placeholder.markdown(full_response)    
    #st.write(repr(result['source_documents'][0].metadata['page']))
    #st.write(repr(result['source_documents'][0]))

    print('5')
    print(time.time())

    def generate_pdf():
        page_number = int(result['source_documents'][0].metadata['page'])
        doc = fitz.open(str(result['source_documents'][0].metadata['source']))

        text = str(result['source_documents'][0].page_content)
        if text != '':
            for page in doc:
                ### SEARCH
                text_instances = page.search_for(text)
    
                ### HIGHLIGHT
                for inst in text_instances:
                    highlight = page.add_highlight_annot(inst)
                    highlight.update()
    
        ### OUTPUT
        doc.save("/home/user/app/pdf2image/output.pdf", garbage=4, deflate=True, clean=True)
    
        # pdf_to_open = repr(result['source_documents'][0].metadata['source'])
    
        def pdf_page_to_image(pdf_file, page_number, output_image):
            # Open the PDF file
            pdf_document = fitz.open(pdf_file)
    
            # Get the specific page
            page = pdf_document[page_number]
    
            # Define the image DPI (dots per inch)
            dpi = 300  # You can adjust this as needed
    
            # Convert the page to an image
            pix = page.get_pixmap(matrix=fitz.Matrix(dpi / 100, dpi / 100))
    
            # Save the image as a PNG file
            pix.save(output_image, "png")
    
            # Close the PDF file
            pdf_document.close()
    
    
        pdf_page_to_image('/home/user/app/pdf2image/output.pdf', page_number, '/home/user/app/pdf2image/output.png')
    
        image = Image.open('/home/user/app/pdf2image/output.png')
        #st.sidebar.image(image)
        st.session_state.image_displayed = True
        return image

    def generate_audio():
        sound_file = BytesIO()
        tts = gTTS(result['result'], lang='en')
        tts.write_to_fp(sound_file)
        #st.sidebar.audio(sound_file)
        st.session_state.sound_played = True
        return sound_file


    #st.button(':speaker:', type='primary',on_click=generate_audio)
    #st.button('Reference',type='primary',on_click=generate_pdf)

    # Create placeholders for output
    image_output = st.empty()
    sound_output = st.empty()
    
    # Create a button to display the image
    # if st.button("Reference"):
    #     image_output.clear()
    #     generate_pdf()
    
    # # Create a button to play the sound
    # if st.button(":speaker:"):
    #     sound_output.clear()
    #     generate_audio()


    # on_audio = st.checkbox(':speaker:', key="speaker")
    # on_ref = st.checkbox('Reference', key="reference")
    # if on_audio:
    #     generate_audio()

    # if on_ref:
    #     generate_pdf()

        # Initialize session state variables
    if "image_displayed" not in st.session_state:
        st.session_state.image_displayed = False
    if "sound_played" not in st.session_state:
        st.session_state.sound_played = False
    
    # Define functions to display the image and play the sound
    # def display_image():
    #     st.image("image.png")
    #     st.session_state.image_displayed = True
    
    # def play_sound():
    #     st.audio("sound.mp3")
    #     st.session_state.sound_played = True
    
    # Create the two buttons
    #st.button("Display Image", on_click=generate_pdf)
    #st.button("Play Sound", on_click=generate_audio)


    
    # # Check if the image has been displayed and display it if it has not
    # if not st.session_state.image_displayed:
    #     generate_pdf()
    
    # # Check if the sound has been played and play it if it has not
    # if not st.session_state.sound_played:
    #     generate_audio()


    for item in output:
        full_response += item
        message_placeholder.markdown(full_response + "▌")
        message_placeholder.markdown(full_response)

    if st.sidebar.button("Display Image"):
        a=generate_pdf()
        message_placeholder.image(a)


    if st.sidebar.button("Play Sound"):
        x=generate_audio()
        message_placeholder.audio(x)
    
    st.session_state.messages.append({"role": "assistant", "content": full_response})