app.py

import os
import streamlit as st
import chatbot as demo_chat
from transformers import AutoModelForCausalLM, AutoTokenizer
from langchain.schema import (
    HumanMessage,
    SystemMessage,
)
from langchain_community.chat_models.huggingface import ChatHuggingFace
from transformers import pipeline

st.title("Hi, I am Chatbot Philio :mermaid:")
st.write("I am your hotel booking assistant for today.")

# tokenizer = AutoTokenizer.from_pretrained("KvrParaskevi/Hotel-Assistant-Attempt4-Llama-2-7b")

tokenizer, model = demo_chat.load_model()

model_identifier = "KvrParaskevi/Hotel-Assistant-Attempt4-Llama-2-7b"
task = "text-generation"  # Change this to your model's task

# Load the model using the pipeline
model_pipeline = pipeline(task, model=model,tokenizer=tokenizer)


#Application 
with st.container():
    st.markdown('<div class="scrollable-div">', unsafe_allow_html=True)
    #Langchain memory in session cache 
    if 'memory' not in st.session_state:
        st.session_state.memory = demo_chat.demo_miny_memory(model)

    #Check if chat history exists in this session
    if 'chat_history' not in st.session_state:
        st.session_state.chat_history = [
            {
                "role": "system",
                "content": "You are a friendly chatbot who always helps the user book a hotel room based on his/her needs."
                + "Based on the current social norms you wait for the user's response to your proposals.",
            },
            {"role": "assistant", "content": "Hello, how can I help you today?"},
        ] #Initialize chat history

    if 'model' not in st.session_state:
        st.session_state.model = model

    #renders chat history
    for message in st.session_state.chat_history:
        if(message["role"]!= "system"):
            with st.chat_message(message["role"]):
                st.write(message["content"])
            
    
    #Set up input text field
    input_text = st.chat_input(placeholder="Here you can chat with our hotel booking model.")

    if input_text:
        with st.chat_message("user"):
            #st.write(input_text)
            st.session_state.chat_history.append({"role" : "user", "content" : input_text}) #append message to chat history

        #chat_response = demo_chat.demo_chain(input_text=input_text, memory=st.session_state.memory, model= chat_model)
        #first_answer = chat_response.split("Human")[0] #Because of Predict it prints the whole conversation.Here we seperate the first answer only.
        tokenized_chat = tokenizer.apply_chat_template(st.session_state.chat_history, tokenize=True, add_generation_prompt=True, return_tensors="pt")
        #st.write(tokenizer.decode(tokenized_chat[0]))
        outputs = model.generate(tokenized_chat, max_new_tokens=128) 
        first_answer = tokenizer.decode(outputs[0][tokenized_chat.shape[1]:],skip_special_tokens=True)
        
        with st.chat_message("assistant"):
            #st.write(first_answer)
            st.session_state.chat_history.append({"role": "assistant", "content": first_answer})
    st.markdown('</div>', unsafe_allow_html=True)