app.py

import os
import openai
import gradio as gr
from langchain.chains import ConversationalRetrievalChain
from langchain.text_splitter import CharacterTextSplitter
from langchain_community.document_loaders import PyMuPDFLoader, PyPDFLoader
from langchain.vectorstores import Chroma
from langchain_community.embeddings import OpenAIEmbeddings
from langchain_community.chat_models import ChatOpenAI
import shutil  # 用於文件複製

# 獲取 OpenAI API 密鑰（初始不使用固定密鑰）
api_key_env = os.getenv("OPENAI_API_KEY")
if api_key_env:
    openai.api_key = api_key_env
else:
    print("未設置固定的 OpenAI API 密鑰。將使用使用者提供的密鑰。")

# 確保向量資料庫目錄存在且有寫入權限
VECTORDB_DIR = os.path.abspath("./data")
os.makedirs(VECTORDB_DIR, exist_ok=True)
os.chmod(VECTORDB_DIR, 0o755)

# 定義載入和處理 PDF 文件的函數
def load_and_process_documents(file_paths, loader_type='PyMuPDFLoader', api_key=None):
    if not api_key:
        raise ValueError("未提供 OpenAI API 密鑰。")
    documents = []

    for file_path in file_paths:
        if not os.path.exists(file_path):
            continue
        try:
            if loader_type == 'PyMuPDFLoader':
                loader = PyMuPDFLoader(file_path)
            elif loader_type == 'PyPDFLoader':
                loader = PyPDFLoader(file_path)
            else:
                continue
            loaded_docs = loader.load()
            if loaded_docs:
                documents.extend(loaded_docs)
        except Exception as e:
            continue

    if not documents:
        raise ValueError("沒有找到任何 PDF 文件或 PDF 文件無法載入。")

    # 分割長文本
    text_splitter = CharacterTextSplitter(chunk_size=1000, chunk_overlap=50)
    documents = text_splitter.split_documents(documents)

    if not documents:
        raise ValueError("分割後的文檔列表為空。請檢查 PDF 文件內容。")

    # 初始化向量資料庫
    try:
        embeddings = OpenAIEmbeddings(openai_api_key=api_key)  # 使用使用者的 API 密鑰
    except Exception as e:
        raise ValueError(f"初始化 OpenAIEmbeddings 時出現錯誤: {e}")

    try:
        vectordb = Chroma.from_documents(
            documents,
            embedding=embeddings,
            persist_directory=VECTORDB_DIR
        )
    except Exception as e:
        raise ValueError(f"初始化 Chroma 向量資料庫時出現錯誤: {e}")

    return vectordb

# 定義聊天處理函數
def handle_query(user_message, chat_history, vectordb, api_key):
    try:
        if not user_message:
            return chat_history

        # 添加角色指令前綴
        preface = """
Instruction: Answer in Traditional Chinese, within 200 characters.
If the question is unrelated to the documents, respond with: 此事無可奉告,話說這件事須請教海虔王...
"""
        query = f"{preface} 查詢內容：{user_message}"

        # 初始化 ConversationalRetrievalChain，並傳遞 openai_api_key
        pdf_qa = ConversationalRetrievalChain.from_llm(
            ChatOpenAI(temperature=0.7, model="gpt-4", openai_api_key=api_key),
            retriever=vectordb.as_retriever(search_kwargs={'k': 6}),
            return_source_documents=True
        )

        # 呼叫模型並處理查詢
        result = pdf_qa.invoke({"question": query, "chat_history": chat_history})

        if "answer" in result:
            chat_history = chat_history + [(user_message, result["answer"])]
        else:
            chat_history = chat_history + [(user_message, "抱歉，未能獲得有效回應。")] 
        return chat_history

    except Exception as e:
        return chat_history + [("系統", f"出現錯誤: {str(e)}")]

# 定義保存 API 密鑰的函數
def save_api_key(api_key, state):
    if not api_key.startswith("sk-"):
        return "請輸入有效的 OpenAI API 密鑰。", state
    state['api_key'] = api_key
    return "API 密鑰已成功保存。您現在可以上傳 PDF 文件並開始提問。", state

# 定義 Gradio 的處理函數
def process_files(files, state):
    if files:
        try:
            api_key = state.get('api_key', None)
            if not api_key:
                return "請先輸入並保存您的 OpenAI API 密鑰。", state

            saved_file_paths = []
            for idx, file_data in enumerate(files):
                filename = f"uploaded_{idx}.pdf"
                save_path = os.path.join(VECTORDB_DIR, filename)
                with open(save_path, "wb") as f:
                    f.write(file_data)
                saved_file_paths.append(save_path)
            vectordb = load_and_process_documents(saved_file_paths, loader_type='PyMuPDFLoader', api_key=api_key)
            state['vectordb'] = vectordb
            return "PDF 文件已成功上傳並處理。您現在可以開始提問。", state
        except Exception as e:
            return f"處理文件時出現錯誤: {e}", state
    else:
        return "請上傳至少一個 PDF 文件。", state

def chat_interface(user_message, chat_history, state):
    vectordb = state.get('vectordb', None)
    api_key = state.get('api_key', None)
    if not vectordb:
        return chat_history, state, "請先上傳 PDF 文件以進行處理。"
    if not api_key:
        return chat_history, state, "請先輸入並保存您的 OpenAI API 密鑰。"

    updated_history = handle_query(user_message, chat_history, vectordb, api_key)
    return updated_history, state, ""

# 設計 Gradio 介面
with gr.Blocks(css="body { background-color: #EBD6D6; }") as demo:
    gr.Markdown("<h1 style='text-align: center;'>AI論壇助理</h1>")

    state = gr.State({"vectordb": None, "api_key": None})

    # API 密鑰輸入框
    api_key_input = gr.Textbox(
        label="輸入您的 OpenAI API 密鑰",
        placeholder="sk-...",
        type="password",
        interactive=True
    )
    save_api_key_btn = gr.Button("保存 API 密鑰")
    api_key_status = gr.Textbox(label="狀態", interactive=False)

    # 上傳 PDF 文件
    gr.Markdown("<span style='font-size: 1.5em; font-weight: bold;'>請上傳AI論壇相關文檔，提供AI相關問題解答</span>")
    upload = gr.File(
        file_count="multiple",
        file_types=[".pdf"],
        label="上傳AI論壇 PDF 文件",
        interactive=True,
        type="binary"
    )
    upload_btn = gr.Button("上傳並處理")
    upload_status = gr.Textbox(label="上傳狀態", interactive=False)

    # 智能諮詢
    gr.Markdown("### AI論壇助理")
    chatbot = gr.Chatbot()

    txt = gr.Textbox(show_label=False, placeholder="請輸入您的AI問題...")
    submit_btn = gr.Button("提問")

    # 綁定事件
    save_api_key_btn.click(
        save_api_key,
        inputs=[api_key_input, state],
        outputs=[api_key_status, state]
    )

    upload_btn.click(
        process_files,
        inputs=[upload, state],
        outputs=[upload_status, state]
    )

    submit_btn.click(
        chat_interface,
        inputs=[txt, chatbot, state],
        outputs=[chatbot, state, txt]
    )

    txt.submit(
        chat_interface,
        inputs=[txt, chatbot, state],
        outputs=[chatbot, state, txt]
    )

# 啟動 Gradio 應用
demo.launch()