app.py

import gradio as gr
import spaces
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
from langchain.llms.base import LLM
from langchain.memory import ConversationBufferMemory
from langchain.chains import LLMChain, ConversationChain
from langchain_community.llms.huggingface_pipeline import HuggingFacePipeline
from langchain.prompts import PromptTemplate, ChatPromptTemplate

@spaces.GPU
def initialize_model_and_tokenizer(model_name="KvrParaskevi/Llama-2-7b-Hotel-Booking-Model"):
    model = AutoModelForCausalLM.from_pretrained(model_name)
    tokenizer = AutoTokenizer.from_pretrained(model_name)
    return model, tokenizer

@spaces.GPU
def load_pipeline():
    model, tokenizer = initialize_model_and_tokenizer()
    pipe = pipeline("text-generation", 
                    model= model, 
                    tokenizer = tokenizer,
                    max_new_tokens = 50,
                    top_k = 30,
                    top_p = 0.7,
                    early_stopping=True,
                    num_beams = 2,
                    temperature = 0.1,
                    repetition_penalty = 1.03)

    llm = HuggingFacePipeline(pipeline = pipe)
    return llm

def chat_interface(inputs):
    question = inputs
    chat_history_tuples = []
    for message in chat_history:
        chat_history_tuples.append((message[0], message[1]))
    
    #result = llm_chain({"input": query, "history": chat_history_tuples})
    result = llm_chain.invoke({
        "input" : inputs,
        "history" : chat_history
    }, return_only_outputs=True)
    formated_result = result['response'] #.split('AI:')[-1]
    return formated_result #.strip()

llm = load_pipeline()
chat_history = []

def chat_output(inputs):
    output = chat_interface(inputs)
    answer_after_input = output.split('AI:')
    answer = answer_after_input[1].split('Human:')
    chat_history.append((inputs, answer[0]))
    return chat_history

template = """You are an AI having conversation with a human. 
Make sure you receive a logical answer from the user from every question to complete the hotel 
booking process.
Current conversation:
{history}
Human: {input}
AI:"""
prompt = PromptTemplate(template=template, input_variables=["history", "input"])

memory = ConversationBufferMemory(memory_key="history", llm = llm)
llm_chain = ConversationChain(llm=llm, memory = memory, prompt= prompt)

with gr.Blocks() as demo:
    #gr.Markdown("Hotel Booking Assistant Chat 🤗")
    #chatbot = gr.Chatbot( [ ],label="Chat history", bubble_full_width=False)
    #message = gr.Textbox(label="Ask me a question!")
    #clear = gr.Button("Clear")
    #llm_chain, llm = init_chain(model, tokenizer)
    chatbot_component = gr.Chatbot(height=300, label = "history")
    textbox_component = gr.Textbox(placeholder="Can I help you to book a hotel?", container=False, label = "input", scale=7)

    demo.chatbot_interface = gr.Interface(
        fn=chat_output, 
        inputs=[
            textbox_component
        ], 
        outputs=chatbot_component,
        title = "Hotel Booking Assistant Chat 🤗",
        description = "I am your hotel booking assistant. Feel free to start chatting with me."
    )
demo.launch()