# 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 import logging # 設置日誌配置 logging.basicConfig(level=logging.INFO) logger = logging.getLogger(__name__) # 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 logger.info("OpenAI API 密鑰已設置。") else: logger.info("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) # 設置適當的權限 logger.info(f"VECTORDB_DIR set to: {VECTORDB_DIR}") # 定義測試 PDF 加載器的函數 def test_pdf_loader(file_path, loader_type='PyMuPDFLoader'): logger.info(f"Testing PDF loader ({loader_type}) with file: {file_path}") try: if loader_type == 'PyMuPDFLoader': loader = PyMuPDFLoader(file_path) elif loader_type == 'PyPDFLoader': loader = PyPDFLoader(file_path) else: logger.error(f"Unknown loader type: {loader_type}") return loaded_docs = loader.load() if loaded_docs: logger.info(f"Successfully loaded {file_path} with {len(loaded_docs)} documents.") logger.info(f"Document content (first 500 chars): {loaded_docs[0].page_content[:500]}") else: logger.error(f"No documents loaded from {file_path}.") except Exception as e: logger.error(f"Error loading {file_path} with {loader_type}: {e}") # 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 = [] logger.info("開始載入上傳的 PDF 文件。") for file_path in file_paths: logger.info(f"載入 PDF 文件: {file_path}") if not os.path.exists(file_path): logger.error(f"文件不存在: {file_path}") continue try: if loader_type == 'PyMuPDFLoader': loader = PyMuPDFLoader(file_path) elif loader_type == 'PyPDFLoader': loader = PyPDFLoader(file_path) else: logger.error(f"Unknown loader type: {loader_type}") continue loaded_docs = loader.load() if loaded_docs: logger.info(f"載入 {file_path} 成功,包含 {len(loaded_docs)} 個文檔。") # 打印第一個文檔的部分內容以確認 logger.info(f"第一個文檔內容: {loaded_docs[0].page_content[:500]}") documents.extend(loaded_docs) else: logger.error(f"載入 {file_path} 但未找到任何文檔。") except Exception as e: logger.error(f"載入 {file_path} 時出現錯誤: {e}") if not documents: raise ValueError("沒有找到任何 PDF 文件或 PDF 文件無法載入。") else: logger.info(f"總共載入了 {len(documents)} 個文檔。") # 分割長文本 text_splitter = CharacterTextSplitter(chunk_size=1000, chunk_overlap=50) documents = text_splitter.split_documents(documents) logger.info(f"分割後的文檔數量: {len(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 logger.info("初始化 OpenAIEmbeddings 成功。") 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 ) logger.info("初始化 Chroma 向量資料庫成功。") 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 = """ 指令: 以繁體中文回答問題,200字以內。你是一位專業心理學家與調酒師,專精於 MBTI 人格與經典調酒主題。 非相關問題,請回應:「目前僅支援 MBTI 分析與經典調酒主題。」。 """ 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}) # 檢查結果並更新聊天歷史 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: logger.error(f"Error in handle_query: {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 logger.info("使用者已保存自己的 OpenAI API 密鑰。") 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): logger.info("process_files called") if files: try: # 檢查是否已保存 API 密鑰 api_key = state.get('api_key', None) if not api_key: logger.error("使用者未提供 OpenAI API 密鑰。") return "請先在「設定 API 密鑰」標籤中輸入並保存您的 OpenAI API 密鑰。", state logger.info(f"Received {len(files)} files") saved_file_paths = [] for idx, file_data in enumerate(files): # 為每個文件分配唯一的文件名 filename = f"uploaded_{idx}.pdf" save_path = os.path.join(VECTORDB_DIR, filename) with open(save_path, "wb") as f: f.write(file_data) # 確認文件是否存在並檢查文件大小 if os.path.exists(save_path): file_size = os.path.getsize(save_path) if file_size > 0: logger.info(f"File successfully saved to: {save_path} (Size: {file_size} bytes)") else: logger.error(f"File saved to {save_path} is empty.") raise ValueError(f"上傳的文件 {filename} 為空。") else: logger.error(f"Failed to save file to: {save_path}") raise FileNotFoundError(f"無法保存文件到 {save_path}") saved_file_paths.append(save_path) # 測試 PDF 加載器,先用 PyMuPDFLoader,再用 PyPDFLoader try: test_pdf_loader(save_path, loader_type='PyMuPDFLoader') except Exception as e: logger.error(f"PyMuPDFLoader failed: {e}") logger.info("Attempting to load with PyPDFLoader...") test_pdf_loader(save_path, loader_type='PyPDFLoader') # 列出 VECTORDB_DIR 中的所有文件 saved_files = os.listdir(VECTORDB_DIR) logger.info(f"Files in VECTORDB_DIR ({VECTORDB_DIR}): {saved_files}") # 列出文件大小 file_sizes = {file: os.path.getsize(os.path.join(VECTORDB_DIR, file)) for file in saved_files} logger.info(f"File sizes in VECTORDB_DIR: {file_sizes}") 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: logger.error(f"Error in process_files: {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, "請先在「設定 API 密鑰」標籤中輸入並保存您的 OpenAI API 密鑰。" # 處理查詢 updated_history = handle_query(user_message, chat_history, vectordb, api_key) return updated_history, state, "" # 設計 Gradio 介面 with gr.Blocks() as demo: gr.Markdown("

MBTI 與經典調酒 AI 助理

") # 定義共享的 state state = gr.State({"vectordb": None, "api_key": None}) with gr.Tab("設定 API 密鑰"): with gr.Row(): with gr.Column(scale=1): 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) with gr.Tab("上傳 PDF 文件"): with gr.Row(): with gr.Column(scale=1): upload = gr.File( file_count="multiple", file_types=[".pdf"], label="上傳 PDF 文件", interactive=True, type="binary" # 修改為 'binary' ) upload_btn = gr.Button("上傳並處理") upload_status = gr.Textbox(label="上傳狀態", interactive=False) with gr.Tab("聊天機器人"): chatbot = gr.Chatbot() with gr.Row(): with gr.Column(scale=0.85): txt = gr.Textbox(show_label=False, placeholder="請輸入您的問題...") with gr.Column(scale=0.15, min_width=0): submit_btn = gr.Button("提問") # 綁定提問按鈕 submit_btn.click( chat_interface, inputs=[txt, chatbot, state], outputs=[chatbot, state, txt] ) # 綁定輸入框的提交事件 txt.submit( chat_interface, inputs=[txt, chatbot, state], outputs=[chatbot, state, txt] ) # 綁定保存 API 密鑰按鈕 save_api_key_btn.click( save_api_key, inputs=[api_key_input, state], outputs=[api_key_status, state] ) # 綁定上傳按鈕 upload_btn.click( process_files, inputs=[upload, state], outputs=[upload_status, state] ) # 啟動 Gradio 應用 demo.launch()