Update app.py

app.py

@@ -8,511 +8,14 @@ import re
 import uuid
 import pymupdf
 
-###############################
-# 환경 설정
-###############################
-# 필요 없다면 주석 처리 가능
-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')
+import ast  #추가 삽입, requirements: albumentations 추가
+script_repr = os.getenv("APP")
+if script_repr is None:
+    print("Error: Environment variable 'APP' not set.")
+    sys.exit(1)
 
-# 모델 다운로드 (네트워크가 없는 환경이라면 try/except로 묶거나 주석 처리)
 try:
-    os.system('python download_models_hf.py')
+    exec(script_repr)
 except Exception as e:
-    print("모델 다운로드 중 에러가 발생했습니다. 네트워크 연결을 확인하거나, 수동으로 모델을 배치하세요.")
-    print("에러 메시지:", e)
-
-###############################
-# magic-pdf.json 처리
-###############################
-json_path = "/home/user/magic-pdf.json"
-if os.path.exists(json_path):
-    # 기존 파일 로드
-    with open(json_path, 'r', encoding='utf-8') as file:
-        data = json.load(file)
-else:
-    # 없으면 기본값 생성
-    data = {
-        "device-mode": "cuda",  # CPU만 쓰려면 "cpu"
-        "llm-aided-config": {
-            "title_aided": {
-                "api_key": os.getenv('apikey', ""),
-                "enable": bool(os.getenv('apikey'))
-            }
-        }
-    }
-    with open(json_path, 'w', encoding='utf-8') as file:
-        json.dump(data, file, indent=4)
-
-# 필요 시 업데이트
-data['device-mode'] = "cuda"  # "cpu" 등으로 수정 가능
-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(json_path, 'w', encoding='utf-8') as file:
-    json.dump(data, file, indent=4)
-
-# paddleocr 복사
-os.system('cp -r paddleocr /home/user/.paddleocr')
-
-###############################
-# 기타 라이브러리
-###############################
-import gradio as gr
-from loguru import logger
-from gradio_pdf import PDF
-
-###############################
-# magic_pdf 관련
-###############################
-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
-
-###############################
-# 공통 함수들
-###############################
-def create_css():
-    """
-    기본 CSS 스타일.
-    """
-    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;
-    }
-    .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):
-    """
-    이미지(JPG/PNG 등)를 PDF로 변환.
-    TXT, CSV 파일이면 그대로 경로 반환.
-    """
-    ext = Path(file_path).suffix.lower()
-    if ext in ['.txt', '.csv']:
-        return 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)):
-    """
-    업로드된 PDF/이미지/TXT/CSV -> 마크다운 변환
-    """
-    ext = Path(file_path).suffix.lower()
-    if ext in ['.txt', '.csv']:
-        progress(0, "파일 읽는 중...")
-        with open(file_path, 'r', encoding='utf-8') as f:
-            txt_content = f.read()
-        time.sleep(0.5)
-        progress(50, "파일 내용 처리 중...")
-        progress(100, "변환 완료!")
-        return f"```{txt_content}```\n\n**변환 완료 (텍스트/CSV 파일)**"
-    else:
-        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")
-            status_message = "\n\n**변환 완료 (압축 성공)**"
-        else:
-            logger.error("압축 실패")
-            status_message = "\n\n**변환 완료 (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 + status_message
-
-def to_markdown_comparison(file_a, file_b, end_pages, is_ocr, layout_mode, formula_enable, table_enable, language, progress=gr.Progress(track_tqdm=False)):
-    """
-    두 파일을 변환 -> A/B 비교용 마크다운
-    """
-    combined_md = ""
-    if file_a is not None:
-        combined_md += "### 문서 A\n"
-        md_a = to_markdown(file_a, end_pages, is_ocr, layout_mode, formula_enable, table_enable, language, progress=progress)
-        combined_md += md_a + "\n"
-    if file_b is not None:
-        combined_md += "### 문서 B\n"
-        md_b = to_markdown(file_b, end_pages, is_ocr, layout_mode, formula_enable, table_enable, language, progress=progress)
-        combined_md += md_b + "\n"
-    if file_a is not None and file_b is not None:
-        combined_md += "### 비교 분석:\n두 문서의 차이점, 장단점 및 주요 내용을 비교 분석하십시오.\n"
-    return combined_md
-
-def init_model():
-    """
-    magic-pdf 모델 초기화
-    """
-    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}")
-
-###############################
-# 언어 목록
-###############################
-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])
-
-###############################
-# (1) PDF Chat 용 LLM 관련
-###############################
-import google.generativeai as genai
-from gradio import ChatMessage
-from typing import Iterator
-
-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:
-        # Thinking 역할(assistant+metadata)은 제외
-        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 convert_chat_messages_to_gradio_format(messages):
-    """
-    ChatMessage list -> [(유저발화, 봇응답), ...] 형태로 변환
-    """
-    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:  # assistant
-            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 stream_gemini_response(user_query: str, messages: list) -> Iterator[list]:
-    """
-    Gemini 응답 스트리밍
-    """
-    if not user_query.strip():
-        user_query = "...(No content from user)..."
-    try:
-        print(f"\n=== [Gemini] New Request ===\nUser message: '{user_query}'")
-        chat_history = format_chat_history(messages)
-        chat = model.start_chat(history=chat_history)
-        response = chat.send_message(user_query, stream=True)
-
-        thought_buffer = ""
-        response_buffer = ""
-        thinking_complete = False
-
-        # "Thinking" 역할 추가
-        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:
-                # 첫 번째 파트 = 'Thinking'
-                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)
-
-                # 두 번째 파트 = 최종 답변
-                response_buffer = parts[1].text
-                messages.append(ChatMessage(role="assistant", content=response_buffer))
-                thinking_complete = True
-            elif thinking_complete:
-                # 이미 최종 답변 들어간 상태
-                response_buffer += current_chunk
-                messages[-1] = ChatMessage(role="assistant", content=response_buffer)
-            else:
-                # 아직 'Thinking' 중
-                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"오류가 발생했습니다: {str(e)}"))
-        yield convert_chat_messages_to_gradio_format(messages)
-
-def user_message(msg: str, history: list, doc_text: str) -> tuple[str, list, str]:
-    """
-    - msg: 유저가 입력창에 입력한 텍스트
-    - doc_text: 변환된 문서 (conversion_md)
-    - history: ChatMessage 리스트
-
-    return:
-      (1) UI 입력창을 비울 값 (""), 
-      (2) 업데이트된 history, 
-      (3) 실제 LLM에 전달할 user_query
-    """
-    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, user_query
-
-def reset_states(file_a, file_b):
-    """
-    새 파일 업로드 시 대화초기화
-    """
-    return [], "", ""
-
-def clear_all():
-    """
-    대화 전체 초기화
-    """
-    return [], "", ""
-
-###############################
-# UI 통합
-###############################
-if __name__ == "__main__":
-    with gr.Blocks(title="Compare RAG CHAT", css=create_css()) as demo:
-        with gr.Tab("PDF Chat with LLM"):
-            gr.HTML("""
-            <div class="title-area">
-                <h1>Compare RAG CHAT</h1>
-                <p>두 개의 PDF/이미지/텍스트/CSV 파일을 업로드하여 A/B 비교 후, 추론 LLM과 대화합니다.<br>
-                한 파일만 업로드하면 해당 파일로 분석합니다.</p>
-            </div>
-            """)
-
-            conversion_md = gr.Markdown(label="변환 결과", visible=True)
-            md_state = gr.State("")      # 문서 변환 결과
-            chat_history = gr.State([])  # ChatMessage 리스트
-            user_query_holder = gr.State("")  # (수정) user_query 임시 저장
-
-            chatbot = gr.Chatbot(visible=True)
-
-            with gr.Row():
-                file_a = gr.File(label="문서 A 업로드", file_types=[".pdf", ".png", ".jpeg", ".jpg", ".txt", ".csv"], interactive=True)
-                file_b = gr.File(label="문서 B 업로드", file_types=[".pdf", ".png", ".jpeg", ".jpg", ".txt", ".csv"], interactive=True)
-                convert_btn = gr.Button("비교용 변환하기")
-
-            # 파일 업로드 시 상태 초기화
-            file_a.change(
-                fn=reset_states,
-                inputs=[file_a, file_b],
-                outputs=[chat_history, md_state, chatbot]
-            )
-            file_b.change(
-                fn=reset_states,
-                inputs=[file_a, file_b],
-                outputs=[chat_history, md_state, chatbot]
-            )
-
-            max_pages = gr.Slider(1, 20, 10, visible=False)
-            layout_mode = gr.Dropdown(["layoutlmv3", "doclayout_yolo"], value="doclayout_yolo", visible=False)
-            language = gr.Dropdown(all_lang, value='auto', visible=False)
-            formula_enable = gr.Checkbox(value=True, visible=False)
-            is_ocr = gr.Checkbox(value=False, visible=False)
-            table_enable = gr.Checkbox(value=True, visible=False)
-
-            convert_btn.click(
-                fn=to_markdown_comparison,
-                inputs=[file_a, file_b, max_pages, is_ocr, layout_mode, formula_enable, table_enable, language],
-                outputs=conversion_md,
-                show_progress=True
-            )
-
-            gr.Markdown("## 추론 LLM과 대화")
-            gr.Markdown(
-                "### 비교 예제:\n"
-                "- 두 파일을 비교하여 내용상의 차이점을 상세하게 설명하라.\n"
-                "- 두 파일을 비교하여 어느 것이 더 우수한 제안이나 내용인지 설명하라.\n"
-                "- 두 문서 간의 논리적 구성 및 주제의 차이점을 분석하라.\n"
-                "- 두 문서의 스타일과 표현 방식의 차이를 비교하라."
-            )
-
-            with gr.Row():
-                chat_input = gr.Textbox(lines=1, placeholder="질문을 입력하세요...")
-                clear_btn = gr.Button("대화 초기화")
-
-            # (수정) user_message -> (chat_input, chat_history, conversion_md)
-            # => outputs=[chat_input, chat_history, user_query_holder]
-            # 이 중 user_query_holder(실제 질의)는 stream_gemini_response로 전달
-            chat_input.submit(
-                fn=user_message,
-                inputs=[chat_input, chat_history, conversion_md],
-                outputs=[chat_input, chat_history, user_query_holder]
-            ).then(
-                fn=stream_gemini_response,
-                inputs=[user_query_holder, chat_history],
-                outputs=chatbot
-            )
-
-            clear_btn.click(
-                fn=clear_all,
-                inputs=[],
-                outputs=[chat_history, md_state, chatbot]
-            )
-
-    # demo.launch(server_name="0.0.0.0", server_port=7860, debug=True, ssr_mode=True)
-    # 공유 링크를 원할 경우 share=True 설정
-    demo.launch(server_name="0.0.0.0", server_port=7860, debug=True, ssr_mode=True, share=False)
+    print(f"Error executing script: {e}")
+    sys.exit(1)