# 4/29/2024

# This version added saving chat history to a log file (need persist data from a space to a dataset)
# Updated the GPT model to gpt-4
# Add timestamp and ip address

# 2/23/2024
# This version uses different method in llama index to define llm model
# Removed deprecated classes and replaced with newest dependencies

# Start by setting token and debug mode before starting schedulers
import os
from huggingface_hub import logging, login
login(token=os.environ.get("HF_TOKEN"), write_permission=True)
#logging.set_verbosity_debug()

import openai
import json
import gradio as gr
from openai import OpenAI

# rebuild storage context and load knowledge index
# from llama_index import StorageContext, load_index_from_storage, LLMPredictor, ServiceContext
from llama_index.core import StorageContext
from llama_index.core import load_index_from_storage
from llama_index.llms.openai import OpenAI
#from llama_index.core import Settings

# add datetime and ip to the log file
from datetime import datetime;
import socket;

# access data folder of persistent storage
from pathlib import Path
from huggingface_hub import CommitScheduler
from uuid import uuid4

# generate an unique identifier for the session
session_id = uuid4()

# deprecated
storage_context = StorageContext.from_defaults(persist_dir='./')
# gpt-3.5_turbo is the current default model
llm_predictor = OpenAI(temperature=0.5, model="gpt-4")
#service_context = ServiceContext.from_defaults(llm=llm_predictor)
#index = load_index_from_storage(storage_context, service_context=service_context)
index = load_index_from_storage(storage_context)

class Chatbot:
    def __init__(self, api_key, index):
        self.index = index
        openai.api_key = api_key
        self.chat_history = []

        # set chat history data path in data folder (persistent storage)
        dataset_dir = Path("logs")
        dataset_dir.mkdir(parents=True, exist_ok=True)
        #self.dataset_path = dataset_dir / f"chat_log_{uuid4()}.json"
        self.dataset_path = dataset_dir / f"chat_log_{session_id}.json"
        

        self.scheduler = CommitScheduler(
            repo_id="history_data",
            repo_type="dataset",
            folder_path=dataset_dir,
            path_in_repo="data",
        )
        
    def generate_response(self, user_input):
        query_engine = index.as_query_engine(llm=llm_predictor)
        response = query_engine.query(user_input)
        
        # generate response
        message = {"role": "assistant", "content": response.response}
        
        return message
    
    # do not need this function if use append mode when dump data in file
    #def load_chat_history(self):
    #    try:
    #        with open(self.dataset_path, 'r') as f:
    #            self.chat_history = json.load(f)
    #    except FileNotFoundError:
    #        pass
    
    def append_chat_history(self, user_input, output):
            # create a dictionary for the chat history
        #self.chat_history = []
        dt = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
        #print(dt)
        hostname = socket.gethostname()
        ip = socket.gethostbyname(hostname)
        #print(ip)
        
        #self.chat_history.append({"role": "datetime", "content": dt})
        #self.chat_history.append({"role": "IP", "content": ip})
        #self.chat_history.append({"role": "user", "content": user_input})
        #self.chat_history.append({"role": "assistant", "content": output})
        
        # save the data in dictionary format
        dictionary = {
            "datetime": dt,
            "ip": ip,
            "user": user_input,
            "assistant": output
        }
        self.chat_history.append(dictionary)

    def save_chat_history(self):
        with self.scheduler.lock:
            with self.dataset_path.open("a") as f:
                json.dump(self.chat_history, f)
                f.write("\n")

def create_bot(user_input):
    bot = Chatbot(os.getenv("OPENAI_API_KEY"), index=index)
    #bot.load_chat_history();
    
    if user_input:
         # use moderations endpoint to check input
        client = openai.OpenAI()
        response_mod = client.moderations.create(input=user_input)
        response_dict = response_mod.model_dump()
        flagged = response_dict['results'][0]['flagged']
        #print("Flagged:", flagged)
    
        if not flagged:
            response_bot = bot.generate_response(user_input)
            output = response_bot['content']
        else:
             output = "Invalid request."
        
        bot.append_chat_history(user_input, output)
        bot.save_chat_history()
        
        return output

inputs = gr.components.Textbox(lines=7, label="Ask questions related to the course. For example, when is the due date for Excel Module 9, what is the assignment late policy, how to use NPV function in Excel, etc.")
outputs = gr.components.Textbox(label="Response")

gr.Interface(fn=create_bot, inputs=inputs, outputs=outputs, title="Virtual TA",
             description="This is a prototype of learning assistant designed for MIS 320 online section (Version 2.0). Powered by ChatGPT-4.\nNote: ChatGPT can make mistakes. Consider checking important information.",
             theme="compact").launch(share=True)