app.py

import gradio as gr
import os

from langchain_community.vectorstores import FAISS
from langchain_community.document_loaders import PyPDFLoader
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.chains import ConversationalRetrievalChain
from langchain_community.embeddings import HuggingFaceEmbeddings 
from langchain_community.llms import HuggingFacePipeline
from langchain.memory import ConversationBufferMemory
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline

# Lista de modelos públicos e leves
list_llm = ["EleutherAI/gpt-neo-125m", "distilbert/distilgpt2"]  
list_llm_simple = [os.path.basename(llm) for llm in list_llm]

# Carregar e dividir documento PDF
def load_doc(list_file_path):
    loaders = [PyPDFLoader(x) for x in list_file_path]
    pages = []
    for loader in loaders:
        pages.extend(loader.load())
    text_splitter = RecursiveCharacterTextSplitter(
        chunk_size=512,  # Reduzido para acelerar a busca
        chunk_overlap=32  # Menor sobreposição para menos processamento
    )  
    doc_splits = text_splitter.split_documents(pages)
    return doc_splits

# Criar banco de vetores
def create_db(splits):
    embeddings = HuggingFaceEmbeddings()
    vectordb = FAISS.from_documents(splits, embeddings)
    return vectordb

# Inicializar o chain LLM local
def initialize_llmchain(llm_model, temperature, max_tokens, top_k, vector_db, progress=gr.Progress()):
    tokenizer = AutoTokenizer.from_pretrained(llm_model)
    model = AutoModelForCausalLM.from_pretrained(
        llm_model,
        device_map="auto",  # Usa GPU se disponível
        torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,  # Otimiza para GPU
        trust_remote_code=True
    )

    # Pipeline otimizado
    pipe = pipeline(
        "text-generation",
        model=model,
        tokenizer=tokenizer,
        max_new_tokens=max_tokens,
        temperature=temperature,
        top_k=top_k,
        do_sample=False,  # Greedy decoding para mais velocidade
        repetition_penalty=1.1,
        return_full_text=False
    )

    llm = HuggingFacePipeline(pipeline=pipe)
    
    memory = ConversationBufferMemory(
        memory_key="chat_history",
        output_key="answer",
        return_messages=True
    )

    retriever = vector_db.as_retriever(search_kwargs={"k": 2})  # Reduzir número de documentos retornados
    qa_chain = ConversationalRetrievalChain.from_llm(
        llm,
        retriever=retriever,
        chain_type="stuff", 
        memory=memory,
        return_source_documents=True,
        verbose=False,
    )
    return qa_chain

# Inicializar banco de dados
def initialize_database(list_file_obj, progress=gr.Progress()):
    list_file_path = [x.name for x in list_file_obj if x is not None]
    doc_splits = load_doc(list_file_path)
    vector_db = create_db(doc_splits)
    return vector_db, "Database created!"

# Inicializar LLM
def initialize_LLM(llm_option, llm_temperature, max_tokens, top_k, vector_db, progress=gr.Progress()):
    llm_name = list_llm[llm_option]
    print("llm_name: ", llm_name)
    qa_chain = initialize_llmchain(llm_name, llm_temperature, max_tokens, top_k, vector_db, progress)
    return qa_chain, "QA chain initialized. Chatbot is ready!"

def format_chat_history(message, chat_history):
    formatted_chat_history = []
    for user_message, bot_message in chat_history:
        formatted_chat_history.append(f"User: {user_message}")
        formatted_chat_history.append(f"Assistant: {bot_message}")
    return formatted_chat_history

def conversation(qa_chain, message, history):
    formatted_chat_history = format_chat_history(message, history)
    response = qa_chain.invoke({"question": message, "chat_history": formatted_chat_history})
    response_answer = response["answer"]
    if response_answer.find("Helpful Answer:") != -1:
        response_answer = response_answer.split("Helpful Answer:")[-1]
    response_sources = response["source_documents"]
    response_source1 = response_sources[0].page_content.strip()
    response_source2 = response_sources[1].page_content.strip()
    response_source3 = ""  # Menos referências para acelerar
    response_source1_page = response_sources[0].metadata["page"] + 1
    response_source2_page = response_sources[1].metadata["page"] + 1
    response_source3_page = 0
    new_history = history + [(message, response_answer)]
    return qa_chain, gr.update(value=""), new_history, response_source1, response_source1_page, response_source2, response_source2_page, response_source3, response_source3_page

def upload_file(file_obj):
    list_file_path = []
    for idx, file in enumerate(file_obj):
        file_path = file_obj.name
        list_file_path.append(file_path)
    return list_file_path

def demo():
    with gr.Blocks(theme=gr.themes.Default(primary_hue="red", secondary_hue="pink", neutral_hue="sky")) as demo:
        vector_db = gr.State()
        qa_chain = gr.State()
        gr.HTML("<center><h1>RAG PDF chatbot</h1><center>")
        gr.Markdown("""<b>Query your PDF documents!</b> This AI agent is designed to perform retrieval augmented generation (RAG) on PDF documents. Optimized for speed without an API token. \
        <b>Please do not upload confidential documents.</b>
        """)
        with gr.Row():
            with gr.Column(scale=86):
                gr.Markdown("<b>Step 1 - Upload PDF documents and Initialize RAG pipeline</b>")
                with gr.Row():
                    document = gr.Files(height=300, file_count="multiple", file_types=["pdf"], interactive=True, label="Upload PDF documents")
                with gr.Row():
                    db_btn = gr.Button("Create vector database")
                with gr.Row():
                    db_progress = gr.Textbox(value="Not initialized", show_label=False)
                gr.Markdown("<style>body { font-size: 16px; }</style><b>Select Large Language Model (LLM) and input parameters</b>")
                with gr.Row():
                    llm_btn = gr.Radio(list_llm_simple, label="Available LLMs", value=list_llm_simple[0], type="index")
                with gr.Row():
                    with gr.Accordion("LLM input parameters", open=False):
                        with gr.Row():
                            slider_temperature = gr.Slider(minimum=0.01, maximum=1.0, value=0.5, step=0.1, label="Temperature", info="Controls randomness (ignored with greedy decoding)", interactive=True)
                        with gr.Row():
                            slider_maxtokens = gr.Slider(minimum=64, maximum=512, value=128, step=64, label="Max New Tokens", info="Maximum number of tokens to be generated", interactive=True)
                        with gr.Row():
                            slider_topk = gr.Slider(minimum=1, maximum=10, value=3, step=1, label="top-k", info="Number of tokens to select (ignored with greedy decoding)", interactive=True)
                with gr.Row():
                    qachain_btn = gr.Button("Initialize Question Answering Chatbot")
                with gr.Row():
                    llm_progress = gr.Textbox(value="Not initialized", show_label=False)

            with gr.Column(scale=200):
                gr.Markdown("<b>Step 2 - Chat with your Document</b>")
                chatbot = gr.Chatbot(height=505)
                with gr.Accordion("Relevant context from the source document", open=False):
                    with gr.Row():
                        doc_source1 = gr.Textbox(label="Reference 1", lines=2, container=True, scale=20)
                        source1_page = gr.Number(label="Page", scale=1)
                    with gr.Row():
                        doc_source2 = gr.Textbox(label="Reference 2", lines=2, container=True, scale=20)
                        source2_page = gr.Number(label="Page", scale=1)
                    with gr.Row():
                        doc_source3 = gr.Textbox(label="Reference 3", lines=2, container=True, scale=20)
                        source3_page = gr.Number(label="Page", scale=1)
                with gr.Row():
                    msg = gr.Textbox(placeholder="Ask a question", container=True)
                with gr.Row():
                    submit_btn = gr.Button("Submit")
                    clear_btn = gr.ClearButton([msg, chatbot], value="Clear")
            
        # Eventos de pré-processamento
        db_btn.click(initialize_database, inputs=[document], outputs=[vector_db, db_progress])
        qachain_btn.click(initialize_LLM, inputs=[llm_btn, slider_temperature, slider_maxtokens, slider_topk, vector_db], outputs=[qa_chain, llm_progress]).then(
            lambda: [None, "", 0, "", 0, "", 0],
            inputs=None,
            outputs=[chatbot, doc_source1, source1_page, doc_source2, source2_page, doc_source3, source3_page],
            queue=False
        )

        # Eventos do chatbot
        msg.submit(conversation, inputs=[qa_chain, msg, chatbot], outputs=[qa_chain, msg, chatbot, doc_source1, source1_page, doc_source2, source2_page, doc_source3, source3_page], queue=False)
        submit_btn.click(conversation, inputs=[qa_chain, msg, chatbot], outputs=[qa_chain, msg, chatbot, doc_source1, source1_page, doc_source2, source2_page, doc_source3, source3_page], queue=False)
        clear_btn.click(lambda: [None, "", 0, "", 0, "", 0], inputs=None, outputs=[chatbot, doc_source1, source1_page, doc_source2, source2_page, doc_source3, source3_page], queue=False)
    
    demo.queue().launch(debug=True)

if __name__ == "__main__":
    demo()