import gradio as gr
from transformers import pipeline, AutoModelForCausalLM, AutoTokenizer
import torch
import theme

from huggingface_hub import from_pretrained_keras
from tensorflow.keras.applications import EfficientNetB0

import tensorflow as tf
from tensorflow import keras
from PIL import Image

theme = theme.Theme()

import os
import sys
sys.path.append('../..')

#langchain
from langchain.text_splitter import RecursiveCharacterTextSplitter, CharacterTextSplitter
from langchain.embeddings import HuggingFaceEmbeddings
from langchain.prompts import PromptTemplate
from langchain.chains import RetrievalQA
from langchain.prompts import ChatPromptTemplate
from langchain.schema import StrOutputParser
from langchain.schema.runnable import Runnable
from langchain.schema.runnable.config import RunnableConfig
from langchain.chains import (
    LLMChain, ConversationalRetrievalChain)
from langchain.vectorstores import Chroma
from langchain.memory import ConversationBufferMemory
from langchain.chains import LLMChain
from langchain.prompts.prompt import PromptTemplate
from langchain.prompts.chat import ChatPromptTemplate, SystemMessagePromptTemplate
from langchain.prompts import SystemMessagePromptTemplate, HumanMessagePromptTemplate, ChatPromptTemplate,  MessagesPlaceholder
from langchain.document_loaders import PyPDFDirectoryLoader
from langchain.output_parsers import PydanticOutputParser
from langchain_community.llms import HuggingFaceHub
from langchain_community.document_loaders import WebBaseLoader
from pydantic.v1 import BaseModel, Field
import shutil


custom_title = "<span style='color: rgb(243, 239, 224);'>Green Greta</span>"


# Cell 2: ChatBot Model
loader = WebBaseLoader(["https://www.epa.gov/recycle/frequent-questions-recycling"])
data=loader.load()
# split documents
text_splitter = RecursiveCharacterTextSplitter(
    chunk_size=1024,
    chunk_overlap=150,
    length_function=len
)
docs = text_splitter.split_documents(data)
# define embedding
embeddings = HuggingFaceEmbeddings(model_name='thenlper/gte-small')
# create vector database from data
persist_directory = 'docs/chroma/'

# Remove old database files if any
shutil.rmtree(persist_directory, ignore_errors=True)
vectordb = Chroma.from_documents(
    documents=docs,
    embedding=embeddings,
    persist_directory=persist_directory
)
# define retriever
retriever = vectordb.as_retriever(search_kwargs={"k": 2}, search_type="mmr")

class FinalAnswer(BaseModel):
    question: str = Field()
    answer: str = Field()

# Assuming you have a parser for the FinalAnswer class
parser = PydanticOutputParser(pydantic_object=FinalAnswer)

template = """
Your name is Greta and you are a recycling chatbot with the objective to anwer questions from user in English or Spanish /
Has sido diseñado y creado por el Grupo 1 del Máster en Data Science & Big Data de la promoción 2023/2024 de la Universidad Complutense de Madrid. Este grupo está fromado por Rocío, María Guillermo, Alejandra, Paloma y Álvaro /
Use the following pieces of context to answer the question /
If the question is English answer in English /
If the question is Spanish answer in Spanish /
Do not mention the word context when you answer a question /
Answer the question fully and provide as much relevant detail as possible. Do not cut your response short /
Context: {context}
User: {question}
{format_instructions}
"""

# Create the chat prompt templates
sys_prompt = SystemMessagePromptTemplate.from_template(template)
qa_prompt = ChatPromptTemplate(
    messages=[
        sys_prompt,
        HumanMessagePromptTemplate.from_template("{question}")],
    partial_variables={"format_instructions": parser.get_format_instructions()}
)
llm = HuggingFaceHub(
    repo_id="mistralai/Mixtral-8x7B-Instruct-v0.1",
    task="text-generation",
    model_kwargs={
        "max_new_tokens": 2000,
        "top_k": 30,
        "temperature": 0.1,
        "repetition_penalty": 1.03
    },
)

qa_chain = ConversationalRetrievalChain.from_llm(
    llm = llm,
    memory = ConversationBufferMemory(llm=llm, memory_key="chat_history", input_key='question', output_key='output'),
    retriever = retriever,
    verbose = True,
    combine_docs_chain_kwargs={'prompt': qa_prompt},
    get_chat_history = lambda h : h,
    rephrase_question = False,
    output_key = 'output',
)

import numpy as np
import soundfile as sf
# Load ASR pipeline
transcriber = pipeline("automatic-speech-recognition", model="openai/whisper-small")

def chat_interface(question, audio_input=None, history=None):
    if audio_input is not None:
        # Function to transcribe the audio input
        def transcribe(audio):
            # Normalize audio
            audio /= np.max(np.abs(audio))

            # Write the audio to a temporary file
            temp_audio_file = "temp_audio.wav"
            sf.write(temp_audio_file, audio, 16000)  # Assuming 16kHz sample rate

            # Transcribe the audio from the temporary file
            return transcriber(temp_audio_file)[0]['transcription']

        # Transcribe the audio input
        question = transcribe(audio_input)


        return question

    # Original chatbot logic
    result = qa_chain.invoke({'question': question})
    output_string = result['output']

    # Find the index of the last occurrence of "answer": in the string
    answer_index = output_string.rfind('"answer":')

    # Extract the substring starting from the "answer": index
    answer_part = output_string[answer_index + len('"answer":'):].strip()

    # Find the next occurrence of a double quote to get the start of the answer value
    quote_index = answer_part.find('"')

    # Extract the answer value between double quotes
    answer_value = answer_part[quote_index + 1:answer_part.find('"', quote_index + 1)]

    return answer_value

chatbot_gradio_app = gr.Interface(
    fn=chat_interface,
    inputs=[
        gr.Textbox(lines=3, label="Type your message here"),
        gr.Audio(label="Record your voice", type='numpy')  # Change type to "microphone"
    ],
    outputs=gr.Textbox(label="Bot's Response"),
)

chatbot_gradio_app.launch()