import streamlit as st
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch
import random

# Load pre-trained DialoGPT-small model and tokenizer
model_name = "microsoft/DialoGPT-small"
model = AutoModelForCausalLM.from_pretrained(model_name)
tokenizer = AutoTokenizer.from_pretrained(model_name)

# Set device to GPU if available for faster inference, otherwise fallback to CPU
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model.to(device)

# Initialize chat history and conversation context
if 'history' not in st.session_state:
    st.session_state['history'] = []
if 'conversation' not in st.session_state:
    st.session_state['conversation'] = []

# Define multiple system prompts to control bot's behavior
system_prompts = [
    "You are a friendly and professional assistant. You respond in a polite and helpful manner.",
    "You are a casual chatbot that likes to engage in fun and interesting conversations, but always stay respectful.",
    "You are a helpful assistant. Your goal is to provide clear and precise answers to any questions.",
    "You are a compassionate and empathetic listener, always responding with kindness and understanding."
]

# Select a random system prompt to start the conversation
def get_system_prompt():
    return random.choice(system_prompts)

def generate_response(input_text):
    # If it's the first interaction, add the system prompt to the conversation history
    if not st.session_state['history']:
        system_prompt = get_system_prompt()
        st.session_state['conversation'].append(f"System: {system_prompt}")
        system_input_ids = tokenizer.encode(system_prompt + tokenizer.eos_token, return_tensors='pt').to(device)
        st.session_state['history'] = system_input_ids[0].tolist()

    # Encode the new user input, add end of string token
    new_user_input_ids = tokenizer.encode(input_text + tokenizer.eos_token, return_tensors='pt').to(device)

    # If there is conversation history, append the new input to it
    if st.session_state['history']:
        history_tensor = torch.tensor(st.session_state['history']).unsqueeze(0).to(device)
        bot_input_ids = torch.cat([history_tensor, new_user_input_ids], dim=-1)
    else:
        bot_input_ids = new_user_input_ids

    # Generate a response from the model
    chat_history_ids = model.generate(bot_input_ids, max_length=1000, pad_token_id=tokenizer.eos_token_id, top_k=50, top_p=0.95, temperature=0.7)

    # Decode the model's output and add it to the history
    chat_history_ids = chat_history_ids[:, bot_input_ids.shape[-1]:]  # only take the latest generated tokens
    bot_output = tokenizer.decode(chat_history_ids[0], skip_special_tokens=True)

    # Update session state history with the new tokens (flattened)
    st.session_state['history'] = chat_history_ids[0].tolist()

    # Add both user input and bot response to the conversation history for display
    st.session_state['conversation'].append(f"You: {input_text}")
    st.session_state['conversation'].append(f"Bot: {bot_output}")

    return bot_output

# Streamlit Interface
st.title("Chat with DialoGPT")

# Display the conversation history
if st.session_state['conversation']:
    for message in st.session_state['conversation']:
        st.markdown(f"<p style='color:gray; padding:5px;'>{message}</p>", unsafe_allow_html=True)

# Create input box for user
user_input = st.text_input("You: ", "")

if user_input:
    # Generate and display the bot's response
    response = generate_response(user_input)
    st.write(f"Bot: {response}")