app-backup.py

import base64
import json
import os
import time
import zipfile
from pathlib import Path
import re
import uuid
import pymupdf

# (기존 magic-pdf 설치 및 설정 로직)
os.system('pip uninstall -y magic-pdf')
os.system('pip install git+https://github.com/opendatalab/MinerU.git@dev')
os.system('wget https://github.com/opendatalab/MinerU/raw/dev/scripts/download_models_hf.py -O download_models_hf.py')
os.system('python download_models_hf.py')

with open('/home/user/magic-pdf.json', 'r') as file:
    data = json.load(file)

data['device-mode'] = "cuda"
if os.getenv('apikey'):
    data['llm-aided-config']['title_aided']['api_key'] = os.getenv('apikey')
    data['llm-aided-config']['title_aided']['enable'] = True

with open('/home/user/magic-pdf.json', 'w') as file:
    json.dump(data, file, indent=4)

os.system('cp -r paddleocr /home/user/.paddleocr')

###############################
# Gradio 및 기타 라이브러리
###############################
import gradio as gr
from loguru import logger

from magic_pdf.data.data_reader_writer import FileBasedDataReader
from magic_pdf.libs.hash_utils import compute_sha256
from magic_pdf.tools.common import do_parse, prepare_env

def create_css():
    """화면 가득 사용 + 스크롤 허용"""
    return """
    .gradio-container {
        width: 100vw !important;
        min-height: 100vh !important;
        margin: 0 !important;
        padding: 0 !important;
        background: linear-gradient(135deg, #EFF6FF 0%, #F5F3FF 100%);
        display: flex;
        flex-direction: column;
        overflow-y: auto !important;
    }
    .title-area {
        text-align: center;
        margin: 1rem auto;
        padding: 1rem;
        background: white;
        border-radius: 1rem;
        box-shadow: 0 4px 6px -1px rgba(0, 0, 0, 0.1);
        max-width: 800px;
    }
    .title-area h1 {
        background: linear-gradient(90deg, #2563EB 0%, #7C3AED 100%);
        -webkit-background-clip: text;
        -webkit-text-fill-color: transparent;
        font-size: 2.5rem;
        font-weight: bold;
        margin-bottom: 0.5rem;
    }
    .title-area p {
        color: #6B7280;
        font-size: 1.1rem;
    }
    .invisible {
        display: none !important;
    }
    .gr-block, .gr-box {
        padding: 0.5rem !important;
    }
    """

def read_fn(path):
    disk_rw = FileBasedDataReader(os.path.dirname(path))
    return disk_rw.read(os.path.basename(path))

def parse_pdf(doc_path, output_dir, end_page_id, is_ocr, layout_mode, formula_enable, table_enable, language):
    os.makedirs(output_dir, exist_ok=True)
    try:
        file_name = f"{str(Path(doc_path).stem)}_{time.time()}"
        pdf_data = read_fn(doc_path)
        parse_method = "ocr" if is_ocr else "auto"
        local_image_dir, local_md_dir = prepare_env(output_dir, file_name, parse_method)
        do_parse(
            output_dir,
            file_name,
            pdf_data,
            [],
            parse_method,
            False,
            end_page_id=end_page_id,
            layout_model=layout_mode,
            formula_enable=formula_enable,
            table_enable=table_enable,
            lang=language,
            f_dump_orig_pdf=False
        )
        return local_md_dir, file_name
    except Exception as e:
        logger.exception(e)

def compress_directory_to_zip(directory_path, output_zip_path):
    try:
        with zipfile.ZipFile(output_zip_path, 'w', zipfile.ZIP_DEFLATED) as zipf:
            for root, dirs, files in os.walk(directory_path):
                for file in files:
                    file_path = os.path.join(root, file)
                    arcname = os.path.relpath(file_path, directory_path)
                    zipf.write(file_path, arcname)
        return 0
    except Exception as e:
        logger.exception(e)
        return -1

def image_to_base64(image_path):
    with open(image_path, "rb") as image_file:
        return base64.b64encode(image_file.read()).decode('utf-8')

def replace_image_with_base64(markdown_text, image_dir_path):
    pattern = r'\!\[(?:[^\]]*)\]\(([^)]+)\)'
    def replace(match):
        relative_path = match.group(1)
        full_path = os.path.join(image_dir_path, relative_path)
        base64_image = image_to_base64(full_path)
        return f"![{relative_path}](data:image/jpeg;base64,{base64_image})"
    return re.sub(pattern, replace, markdown_text)

def to_pdf(file_path):
    with pymupdf.open(file_path) as f:
        if f.is_pdf:
            return file_path
        else:
            pdf_bytes = f.convert_to_pdf()
            unique_filename = f"{uuid.uuid4()}.pdf"
            tmp_file_path = os.path.join(os.path.dirname(file_path), unique_filename)
            with open(tmp_file_path, 'wb') as tmp_pdf_file:
                tmp_pdf_file.write(pdf_bytes)
            return tmp_file_path

def to_markdown(file_path, end_pages, is_ocr, layout_mode, formula_enable, table_enable, language, progress=gr.Progress(track_tqdm=False)):
    """PDF 변환 함수 (프로그레스바 표시)"""
    progress(0, "PDF로 변환 중...")
    file_path = to_pdf(file_path)
    time.sleep(0.5)

    if end_pages > 20:
        end_pages = 20

    progress(20, "문서 파싱 중...")
    local_md_dir, file_name = parse_pdf(file_path, './output', end_pages - 1, is_ocr,
                                        layout_mode, formula_enable, table_enable, language)
    time.sleep(0.5)

    progress(50, "압축(zip) 생성 중...")
    archive_zip_path = os.path.join("./output", compute_sha256(local_md_dir) + ".zip")
    zip_archive_success = compress_directory_to_zip(local_md_dir, archive_zip_path)
    if zip_archive_success == 0:
        logger.info("압축 성공")
    else:
        logger.error("압축 실패")
    time.sleep(0.5)

    progress(70, "마크다운 읽는 중...")
    md_path = os.path.join(local_md_dir, file_name + ".md")
    with open(md_path, 'r', encoding='utf-8') as f:
        txt_content = f.read()
    time.sleep(0.5)

    progress(90, "이미지 base64 변환 중...")
    md_content = replace_image_with_base64(txt_content, local_md_dir)
    time.sleep(0.5)

    progress(100, "변환 완료!")
    return md_content

def init_model():
    from magic_pdf.model.doc_analyze_by_custom_model import ModelSingleton
    try:
        model_manager = ModelSingleton()
        txt_model = model_manager.get_model(False, False)
        logger.info("txt_model init final")
        ocr_model = model_manager.get_model(True, False)
        logger.info("ocr_model init final")
        return 0
    except Exception as e:
        logger.exception(e)
        return -1

model_init = init_model()
logger.info(f"model_init: {model_init}")

latin_lang = [
    'af','az','bs','cs','cy','da','de','es','et','fr','ga','hr','hu','id','is','it','ku',
    'la','lt','lv','mi','ms','mt','nl','no','oc','pi','pl','pt','ro','rs_latin','sk','sl',
    'sq','sv','sw','tl','tr','uz','vi','french','german'
]
arabic_lang = ['ar','fa','ug','ur']
cyrillic_lang = ['ru','rs_cyrillic','be','bg','uk','mn','abq','ady','kbd','ava','dar','inh','che','lbe','lez','tab']
devanagari_lang = ['hi','mr','ne','bh','mai','ang','bho','mah','sck','new','gom','sa','bgc']
other_lang = ['ch','en','korean','japan','chinese_cht','ta','te','ka']

all_lang = ['', 'auto']
all_lang.extend([*other_lang, *latin_lang, *arabic_lang, *cyrillic_lang, *devanagari_lang])

import google.generativeai as genai
from gradio import ChatMessage
from typing import Iterator
import time

GEMINI_API_KEY = os.getenv("GEMINI_API_KEY")
genai.configure(api_key=GEMINI_API_KEY)
model = genai.GenerativeModel("gemini-2.0-flash-thinking-exp-1219")

def format_chat_history(messages: list) -> list:
    """Gemini가 이해할 수 있는 (role, parts[]) 형식으로 변환"""
    formatted_history = []
    for message in messages:
        if not (message.role == "assistant" and hasattr(message, "metadata")):
            formatted_history.append({
                "role": "user" if message.role == "user" else "assistant",
                "parts": [message.content]
            })
    return formatted_history

def stream_gemini_response(user_message: str, messages: list) -> Iterator[list]:
    """Gemini 응답 스트리밍 (user_message가 공백이면 기본 문구로 교체)"""
    if not user_message.strip():
        user_message = "...(No content from user)..."

    try:
        print(f"\n=== [Gemini] New Request ===\nUser message: '{user_message}'")
        chat_history = format_chat_history(messages)
        chat = model.start_chat(history=chat_history)
        response = chat.send_message(user_message, stream=True)

        thought_buffer = ""
        response_buffer = ""
        thinking_complete = False

        messages.append(
            ChatMessage(
                role="assistant",
                content="",
                metadata={"title": "⚙️ Thinking: *The thoughts produced by the model are experimental"}
            )
        )
        yield convert_chat_messages_to_gradio_format(messages)

        for chunk in response:
            parts = chunk.candidates[0].content.parts
            current_chunk = parts[0].text

            if len(parts) == 2 and not thinking_complete:
                thought_buffer += current_chunk
                messages[-1] = ChatMessage(
                    role="assistant",
                    content=thought_buffer,
                    metadata={"title": "⚙️ Thinking: *The thoughts produced by the model are experimental"}
                )
                yield convert_chat_messages_to_gradio_format(messages)

                response_buffer = parts[1].text
                messages.append(ChatMessage(role="assistant", content=response_buffer))
                thinking_complete = True
            elif thinking_complete:
                response_buffer += current_chunk
                messages[-1] = ChatMessage(role="assistant", content=response_buffer)
            else:
                thought_buffer += current_chunk
                messages[-1] = ChatMessage(
                    role="assistant",
                    content=thought_buffer,
                    metadata={"title": "⚙️ Thinking: *The thoughts produced by the model are experimental"}
                )

            yield convert_chat_messages_to_gradio_format(messages)

        print(f"\n=== [Gemini] Final Response ===\n{response_buffer}")

    except Exception as e:
        print(f"\n=== [Gemini] Error ===\n{str(e)}")
        messages.append(ChatMessage(role="assistant", content=f"I encountered an error: {str(e)}"))
        yield convert_chat_messages_to_gradio_format(messages)

def convert_chat_messages_to_gradio_format(messages):
    """ChatMessage list -> [ (유저발화, 봇응답), (...), ... ]"""
    gradio_chat = []
    user_text, assistant_text = None, None

    for msg in messages:
        if msg.role == "user":
            if user_text is not None or assistant_text is not None:
                gradio_chat.append((user_text or "", assistant_text or ""))
            user_text = msg.content
            assistant_text = None
        else:
            if user_text is None:
                user_text = ""
            if assistant_text is None:
                assistant_text = msg.content
            else:
                assistant_text += msg.content

    if user_text is not None or assistant_text is not None:
        gradio_chat.append((user_text or "", assistant_text or ""))

    return gradio_chat

def user_message(msg: str, history: list, doc_text: str) -> tuple[str, list]:
    """doc_text(마크다운) 사용해 질문 자동 변형"""
    if doc_text.strip():
        user_query = f"다음 문서를 참고하여 답변:\n\n{doc_text}\n\n질문: {msg}"
    else:
        user_query = msg

    history.append(ChatMessage(role="user", content=user_query))
    return "", history

def reset_states(_):
    """
    새 파일 업로드 시 
    - chat_history -> 빈 리스트
    - md_state -> 빈 문자열
    - chatbot -> 빈 list of tuples
    """
    return [], "", []

with gr.Blocks(title="VisionOCR", css=create_css()) as demo:
    gr.HTML("""
    <div class="title-area">
        <h1>VisionOCR</h1>
        <p>PDF/이미지 -> 텍스트(마크다운) 변환 후, 추 LLM과 대화</p>
    </div>
    """)

    md_state = gr.State("")      # 변환된 마크다운 텍스트
    chat_history = gr.State([])  # ChatMessage 리스트

    # 업로드 & 변환
    with gr.Row():
        file = gr.File(label="PDF/이미지 업로드", file_types=[".pdf", ".png", ".jpeg", ".jpg"], interactive=True)
        convert_btn = gr.Button("변환하기")

    # 새 파일 업로드 시: 이전 대화/마크다운/챗봇 초기화
    chatbot = gr.Chatbot(height=600)  # 실제 Chatbot 컴포넌트

    file.change(
        fn=reset_states,
        inputs=file,
        outputs=[chat_history, md_state, chatbot]  # <--- 문자열 "chatbot" 아니라 chatbot 컴포넌트 객체
    )

    # 숨긴 컴포넌트
    max_pages = gr.Slider(1,20,10, visible=False, elem_classes="invisible")
    layout_mode = gr.Dropdown(["layoutlmv3","doclayout_yolo"],value="doclayout_yolo",visible=False,elem_classes="invisible")
    language = gr.Dropdown(all_lang, value='auto', visible=False, elem_classes="invisible")
    formula_enable = gr.Checkbox(value=True, visible=False, elem_classes="invisible")
    is_ocr = gr.Checkbox(value=False, visible=False, elem_classes="invisible")
    table_enable = gr.Checkbox(value=True, visible=False, elem_classes="invisible")

    convert_btn.click(
        fn=to_markdown,
        inputs=[file, max_pages, is_ocr, layout_mode, formula_enable, table_enable, language],
        outputs=md_state,
        show_progress=True  # 프로그레스바 표시
    )

    # Gemini Chat
    gr.Markdown("##추론 LLM과 대화")

    with gr.Row():
        chat_input = gr.Textbox(lines=1, placeholder="질문을 입력하세요...")
        clear_btn = gr.Button("대화 초기화")

    chat_input.submit(
        fn=user_message,
        inputs=[chat_input, chat_history, md_state],
        outputs=[chat_input, chat_history]
    ).then(
        fn=stream_gemini_response,
        inputs=[chat_input, chat_history],
        outputs=chatbot
    )

    def clear_all():
        return [], "", []

    clear_btn.click(
        fn=clear_all,
        inputs=[],
        outputs=[chat_history, md_state, chatbot]
    )

if __name__ == "__main__":
    demo.launch(server_name="0.0.0.0", server_port=7860, debug=True)