app.py

import streamlit as st
from langchain.chains.question_answering import load_qa_chain
from langchain import PromptTemplate, LLMChain
from langchain import HuggingFaceHub
from PyPDF2 import PdfReader
from langchain.text_splitter import CharacterTextSplitter
from langchain.document_loaders import TextLoader
import requests
from pathlib import Path
from time import sleep
import torch
import os
import random
import string
from dotenv import load_dotenv
load_dotenv()

st.set_page_config(page_title="CPEG (EN) AI Chat Assistant", layout="wide")
st.subheader("China Patent Examination Guideline (EN) AI Chat Assistant")
#st.write('---')

css_file = "main.css"
with open(css_file) as f:
    st.markdown("<style>{}</style>".format(f.read()), unsafe_allow_html=True)

HUGGINGFACEHUB_API_TOKEN = os.getenv('HUGGINGFACEHUB_API_TOKEN')
model_id = os.getenv('model_id')
hf_token = os.getenv('hf_token')
repo_id = os.getenv('repo_id')
HUGGINGFACEHUB_API_TOKEN = os.environ.get('HUGGINGFACEHUB_API_TOKEN')
model_id = os.environ.get('model_id')
hf_token = os.environ.get('hf_token')
repo_id = os.environ.get('repo_id')

api_url = f"https://api-inference.huggingface.co/pipeline/feature-extraction/{model_id}"
headers = {"Authorization": f"Bearer {hf_token}"}

def get_embeddings(input_str_texts):
    response = requests.post(api_url, headers=headers, json={"inputs": input_str_texts, "options":{"wait_for_model":True}})
    return response.json()

llm = HuggingFaceHub(repo_id=repo_id,
                     model_kwargs={"min_length":100,
                                   "max_new_tokens":1024, "do_sample":True,
                                   "temperature":0.1,
                                   "top_k":50,
                                   "top_p":0.95, "eos_token_id":49155})

chain = load_qa_chain(llm=llm, chain_type="stuff")

def generate_random_string(length):
    letters = string.ascii_lowercase
    return ''.join(random.choice(letters) for i in range(length))  
    
texts=""
initial_embeddings=""
db_embeddings = ""
i_file_path=""
file_path = ""
wechat_image= "WeChatCode.jpg"

st.sidebar.markdown(
    """
    <style>
    .blue-underline {
        text-decoration: bold;
        color: blue;
    }
    </style>
    """,
    unsafe_allow_html=True
)

st.markdown(
    """
    <style>
        [data-testid=stSidebar] [data-testid=stImage]{
            text-align: center;
            display: block;
            margin-left: auto;
            margin-right: auto;
            width: 50%;
        }
    </style>
    """, unsafe_allow_html=True
)

with st.sidebar:
    option = st.sidebar.selectbox("Select the content to Chat:", ("Part I: Preliminary Examination", "Part II: Substantive Examination", "Part III: Examination of International Applications Entering the National Phase", "Part IV: Examination of Requests for Reexamination and for Invalidation", "Part V: Processing of Patent Applications and Procedural Matters", "Index", "Annexes"))
#    if option == "TOC: Table of Contents":
#        file_path = os.path.join(os.getcwd(), "CPEGFullV2010ENContents.pdf")
    if option == "Part I: Preliminary Examination":
        file_path = os.path.join(os.getcwd(), "CPEGFullV2010ENPartI.pdf")
    elif option == "Part II: Substantive Examination":
        file_path = os.path.join(os.getcwd(), "CPEGFullV2010ENPartII.pdf")
    elif option == "Part III: Examination of International Applications Entering the National Phase":
        file_path = os.path.join(os.getcwd(), "CPEGFullV2010ENPartIII.pdf")
    elif option == "Part IV: Examination of Requests for Reexamination and for Invalidation":
        file_path = os.path.join(os.getcwd(), "CPEGFullV2010ENPartIV.pdf")
    elif option == "Part V: Processing of Patent Applications and Procedural Matters":
        file_path = os.path.join(os.getcwd(), "CPEGFullV2010ENPartV.pdf")
    elif option == "Index":
        file_path = os.path.join(os.getcwd(), "CPEGFullV2010ENIndex.pdf")
    elif option == "Annexes":
        file_path = os.path.join(os.getcwd(), "CPEGFullV2010ENAnnexes.pdf")
    else:
        st.write("Choose which part to Chat first.")
        st.stop()
    st.write("Caution: This app is built based on the English Version of CPEG (2010). For most recent version, please refer to the CNIPA official source.")
    st.write("Disclaimer: This app is for information purpose only. NO liability could be claimed against whoever associated with this app in any manner. User should consult a qualified legal professional for legal advice.")
    st.subheader("Enjoy Chatting!")
    st.sidebar.markdown("Contact: [aichat101@foxmail.com](mailto:aichat101@foxmail.com)")
    st.sidebar.markdown('WeChat: <span class="blue-underline">pat2win</span>, or scan the code below.', unsafe_allow_html=True)
    st.image(wechat_image)
    st.sidebar.markdown('<span class="blue-underline">Life Enhancing with AI.</span>', unsafe_allow_html=True)      
    try:        
        with st.spinner("Preparing materials for you..."):
            doc_reader = PdfReader(file_path)
            raw_text = ''
            for i, page in enumerate(doc_reader.pages):
                text = page.extract_text()
                if text:
                    raw_text += text
#            text_splitter = RecursiveCharacterTextSplitter(        
            text_splitter = CharacterTextSplitter(        
                separator = "\n",
                chunk_size = 1000,
                chunk_overlap  = 200, #striding over the text
                length_function = len,
            )
            temp_texts = text_splitter.split_text(raw_text)
            texts = temp_texts
            initial_embeddings=get_embeddings(texts)
            db_embeddings = torch.FloatTensor(initial_embeddings) 
    except Exception as e:
        st.write("Unknow error.")
        print("Unknow error.")
        st.stop()

user_question = st.text_input("Enter your question & query CPEG (EN):")

if user_question !="" and not user_question.strip().isspace() and not user_question == "" and not user_question.strip() == "" and not user_question.isspace():       
    #st.write("Your question: "+user_question)
    print("Your question: "+user_question)
    print()
else:
    print("Please enter your question first.")
    st.stop()
#    st.write("Please enter your question first.")

q_embedding=get_embeddings(user_question)
final_q_embedding = torch.FloatTensor(q_embedding)

from sentence_transformers.util import semantic_search
hits = semantic_search(final_q_embedding, db_embeddings, top_k=5)

for i in range(len(hits[0])):
    print(texts[hits[0][i]['corpus_id']])
    print()

page_contents = []
for i in range(len(hits[0])):
    page_content = texts[hits[0][i]['corpus_id']]
    page_contents.append(page_content)

print(page_contents)
print()
temp_page_contents=str(page_contents)
print()
final_page_contents = temp_page_contents.replace('\\n', '') 
print(final_page_contents)
print()
print("AI Thinking...Please wait a while to Cheers!")
print()

random_string = generate_random_string(20)

with st.spinner("AI Thinking...Please wait a while to Cheers!"):
    i_file_path = random_string + ".txt"
    with open(i_file_path, "w", encoding="utf-8") as file:
        file.write(final_page_contents)
    loader = TextLoader(i_file_path, encoding="utf-8")
    loaded_documents = loader.load()
    temp_ai_response=chain.run(input_documents=loaded_documents, question=user_question)
    final_ai_response=temp_ai_response.partition('<|end|>')[0]
    i_final_ai_response = final_ai_response.replace('\n', '')
    print("AI Response:")
    print(i_final_ai_response)
    print("Have more questions? Go ahead and continue asking your AI assistant : )")
    st.write("AI Response:")
    st.write(i_final_ai_response)
#    st.write("---")
#    st.write("Have more questions? Go ahead and continue asking your AI assistant : )")