app.py

import gradio as gr
import pandas as pd
import json

from langchain.document_loaders import DataFrameLoader
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.llms import HuggingFaceHub
from langchain.embeddings.sentence_transformer import SentenceTransformerEmbeddings
from langchain.vectorstores import Chroma
from langchain.chains import RetrievalQA

from trafilatura import fetch_url, extract
from trafilatura.spider import focused_crawler
from trafilatura.settings import use_config




def loading_website():
    return "Loading..."

def url_changes(url, pages_to_visit, urls_to_scrape, repo_id):
    to_visit, links = focused_crawler(url, max_seen_urls=pages_to_visit, max_known_urls=urls_to_scrape)
    print(f"{len(links)} to be crawled")

    config = use_config()
    config.set("DEFAULT", "EXTRACTION_TIMEOUT", "0")

    results_df = pd.DataFrame()
    for url in links:
        downloaded = fetch_url(url)
        if downloaded:
          result = extract(downloaded, output_format='json', config=config)
          result = json.loads(result)

          results_df = pd.concat([results_df, pd.DataFrame.from_records([result])])
    results_df.to_csv("./data.csv")
    
    df = pd.read_csv("./data.csv")
    loader = DataFrameLoader(df, page_content_column="text")
    documents = loader.load()
    print(f"{len(documents)} documents loaded") 

    text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=200)
    texts = text_splitter.split_documents(documents)
    print(f"documents splitted into {len(texts)} chunks") 
    
    embeddings = SentenceTransformerEmbeddings(model_name="jhgan/ko-sroberta-multitask")

    persist_directory = './vector_db'
    db = Chroma.from_documents(texts, embeddings, persist_directory=persist_directory)
    retriever = db.as_retriever()
    
    llm = HuggingFaceHub(repo_id=repo_id, model_kwargs={"temperature":0.1, "max_new_tokens":250})
    global qa
    qa = RetrievalQA.from_chain_type(llm=llm, chain_type="stuff", retriever=retriever, return_source_documents=True)
    return "Ready"

def add_text(history, text):
    history = history + [(text, None)]
    return history, ""

def bot(history):
    response = infer(history[-1][0])
    history[-1][1] = response['result']
    return history

def infer(question):

    query = question
    result = qa({"query": query})

    return result

css="""
#col-container {max-width: 700px; margin-left: auto; margin-right: auto;}
"""

title = """
<div style="text-align: center;max-width: 700px;">
    <h1>Chat with your website</h1>
    <p style="text-align: center;">Enter target URL, click the "Load website to LangChain" button</p>
</div>
"""


with gr.Blocks(css=css) as demo:
    with gr.Column(elem_id="col-container"):
        gr.HTML(title)

        with gr.Column():
            target_url = gr.Textbox(label="Load URL", placeholder="Enter target URL here. EX: https://www.penta.co.kr/")
            #pdf_doc = gr.File(label="Load URL", file_types=['.pdf'], type="file")
            repo_id = gr.Dropdown(label="LLM", choices=["google/flan-ul2", "OpenAssistant/oasst-sft-1-pythia-12b", "beomi/KoAlpaca-Polyglot-12.8B"], value="google/flan-ul2")
            with gr.Row():
                langchain_status = gr.Textbox(label="Status", placeholder="", interactive=False)
                load_pdf = gr.Button("Load website to langchain")

        chatbot = gr.Chatbot([], elem_id="chatbot").style(height=350)
        question = gr.Textbox(label="Question", placeholder="Type your question and hit Enter ")
        submit_btn = gr.Button("Send message")
    #load_pdf.click(loading_pdf, None, langchain_status, queue=False)
    repo_id.change(url_changes, inputs=[target_url, gr.Number(value=5, visible=False), gr.Number(value=50, visible=False), repo_id], outputs=[langchain_status], queue=False)
    load_pdf.click(url_changes, inputs=[target_url, gr.Number(value=5, visible=False), gr.Number(value=50, visible=False), repo_id], outputs=[langchain_status], queue=False)
    question.submit(add_text, [chatbot, question], [chatbot, question]).then(
        bot, chatbot, chatbot
    )
    submit_btn.click(add_text, [chatbot, question], [chatbot, question]).then(
        bot, chatbot, chatbot
    )

demo.launch()