app.py

import os
from huggingface_hub import InferenceClient
import gradio as gr
import nltk
import torch
from transformers import DistilBertTokenizer, DistilBertModel
from duckduckgo_search import ddg
from langchain.chains import RetrievalQA
from langchain.document_loaders import UnstructuredFileLoader
from langchain.embeddings import HuggingFaceBgeEmbeddings
from langchain.vectorstores import Chroma
from transformers import DistilBertConfig, DistilBertModel

# Initialize tokenizer and model for embedding
tokenizer = DistilBertTokenizer.from_pretrained("distilbert-base-uncased-finetuned-sst-2-english")
embedding_model_name = "distilbert/distilbert-base-uncased-finetuned-sst-2-english"
DEVICE = "cuda" if torch.cuda.is_available() else "mps" if torch.backends.mps.is_available() else "cpu"

# Load Qwen 2 for text generation
qwen_text_gen = InferenceClient("HuggingFaceH4/zephyr-7b-beta")

# Function to search the web
def search_web(query):
    results = ddg(query)
    web_content = ''
    if results:
        for result in results:
            web_content += result['body']
    return web_content

# Function to initialize knowledge vector store
def init_knowledge_vector_store(file):
    if file is None:
        return
    filepath = file.name
    distilbert_embedding = HuggingFaceBgeEmbeddings(model_name=embedding_model_name)
    loader = UnstructuredFileLoader(filepath, mode="elements")
    docs = loader.load()
    Chroma.from_documents(docs, distilbert_embedding, persist_directory="./vector_store")

# Function to get knowledge vector store
def get_knowledge_vector_store():
    distilbert_embedding = HuggingFaceBgeEmbeddings(model_name=embedding_model_name)
    vector_store = Chroma(embedding_function=distilbert_embedding, persist_directory="./vector_store")
    return vector_store

# Function to get knowledge-based answer
def get_knowledge_based_answer(query, qwen_text_gen, vector_store, VECTOR_SEARCH_TOP_K, web_content):
    if web_content:
        prompt_template = f"""Answer the user's question based on the following known information.
                            Known web search content: {web_content} """ + """
                            Known Content:
                            {context}
                            question:
                            {question}"""
    else:
        prompt_template = """Answer the user's question based on the known information.
            Known Content:
            {context}
            question:
            {question}"""
    prompt = PromptTemplate(template=prompt_template, input_variables=["context", "question"])
    
    knowledge_chain = RetrievalQA.from_llm(
        llm=qwen_text_gen, 
        retriever=vector_store.as_retriever(search_kwargs={"k": VECTOR_SEARCH_TOP_K}), 
        prompt=prompt
    )
    
    knowledge_chain.combine_documents_chain.document_prompt = PromptTemplate(
        input_variables=["page_content"], 
        template="{page_content}"
    )
    
    knowledge_chain.return_source_documents = True
    
    result = knowledge_chain.invoke({"query": query})
    
    return result['result']

# Function to clear session
def clear_session():
    return '', None

# Function to predict
def predict(input, qwen_text_gen, VECTOR_SEARCH_TOP_K, use_web, key=None, history=None):
    if history == None:
        history = []
    vector_store = get_knowledge_vector_store()
    if use_web == 'True':
        web_content = search_web(query=input)
        if web_content is None:
            web_content = ""
    else:
        web_content = ''

    resp = get_knowledge_based_answer(
        query=input,
        qwen_text_gen=qwen_text_gen,
        vector_store=vector_store,
        VECTOR_SEARCH_TOP_K=VECTOR_SEARCH_TOP_K,
        web_content=web_content,
    )
    history.append((input, resp))
    return '', history, history

# Gradio interface setup
block = gr.Blocks()
with block as demo:
    gr.Markdown("<h1><center>Chat History </center></h1>")
    with gr.Row():
        with gr.Column(scale=1):
            file = gr.File(label='Please upload txt, md, docx type files', file_types=['.txt', '.md', '.docx'])
            get_vs = gr.Button("Generate Knowledge Base")
            get_vs.click(init_knowledge_vector_store, inputs=[file])
            
            use_web = gr.Radio(["True", "False"], label="Web Search", value="False")
            
            VECTOR_SEARCH_TOP_K = gr.Slider(1, 10, value=5, step=1, label="vector search top k", interactive=True)

        with gr.Column(scale=4):
            chatbot = gr.Chatbot(label='Ming History Knowledge Question and Answer Assistant', height=600)
            message = gr.Textbox(label='Please enter your question')
            state = gr.State()

            with gr.Row():
                clear_history = gr.Button("Clear history conversation")
                send = gr.Button("Send")
                send.click(predict,
                           inputs=[message, qwen_text_gen, VECTOR_SEARCH_TOP_K, use_web, state],
                           outputs=[message, chatbot, state])
                clear_history.click(fn=clear_session, inputs=[], outputs=[chatbot, state], queue=False)
                
                message.submit(predict,
                               inputs=[message, qwen_text_gen, VECTOR_SEARCH_TOP_K, use_web, state],
                               outputs=[message, chatbot, state])

demo.queue().launch(share=False)