import gradio as gr import os import tempfile from typing import List, Tuple, Optional import dashscope from langchain.text_splitter import RecursiveCharacterTextSplitter from langchain_community.embeddings import HuggingFaceEmbeddings from langchain_community.vectorstores import Chroma from langchain_community.document_loaders import PyPDFLoader, Docx2txtLoader, TextLoader import shutil # ============== 配置 ============== DASHSCOPE_API_KEY = os.getenv("DASHSCOPE_API_KEY", "") if DASHSCOPE_API_KEY: dashscope.api_key = DASHSCOPE_API_KEY QWEN_MODEL = "qwen-turbo" # 全局变量 vector_store = None docs_loaded = False # ============== 工具函数 ============== def load_documents(file_paths: List[str]) -> List: """加载各种格式的文档""" documents = [] for file_path in file_paths: try: if file_path.endswith('.pdf'): loader = PyPDFLoader(file_path) elif file_path.endswith('.docx'): loader = Docx2txtLoader(file_path) elif file_path.endswith('.txt'): loader = TextLoader(file_path, encoding='utf-8') else: continue documents.extend(loader.load()) except Exception as e: print(f"加载文件失败 {file_path}: {e}") return documents def create_vector_store(documents: List): """创建向量数据库""" text_splitter = RecursiveCharacterTextSplitter( chunk_size=500, chunk_overlap=50, length_function=len, ) chunks = text_splitter.split_documents(documents) embeddings = HuggingFaceEmbeddings( model_name="sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2", model_kwargs={'device': 'cpu'}, encode_kwargs={'normalize_embeddings': True} ) vector_store = Chroma.from_documents( documents=chunks, embedding=embeddings, persist_directory=None ) return vector_store def chat_with_qwen(messages: List[dict]) -> str: """调用通义千问 API""" try: response = dashscope.Generation.call( model=QWEN_MODEL, messages=messages, result_format='message' ) if response.status_code == 200: return response.output.choices[0].message.content else: return f"API 错误:{response.code} - {response.message}" except Exception as e: return f"调用失败:{str(e)}" def retrieve_context(query: str, top_k: int = 3) -> str: """从向量数据库检索相关内容""" global vector_store if vector_store is None: return "" try: results = vector_store.similarity_search(query, k=top_k) context = "\n\n".join([doc.page_content for doc in results]) return context except Exception as e: return f"检索失败:{str(e)}" # ============== Gradio 界面 ============== with gr.Blocks(title="通义千问混合机器人", theme=gr.themes.Soft()) as demo: gr.Markdown(""" # 🤖 通义千问混合机器人 支持两种模式: - **💬 聊天模式**:自由对话,什么问题都可以问 - **📄 文档问答**:上传文档后,基于文档内容回答问题 **使用前请确保已在 Space 设置中配置 `DASHSCOPE_API_KEY` 环境变量** """) # 状态存储 chat_state = gr.State([]) doc_store = gr.State(None) with gr.Row(): with gr.Column(scale=1): gr.Markdown("### 📁 文档管理") file_upload = gr.File( label="上传文档 (PDF/Word/TXT)", file_types=[".pdf", ".docx", ".txt"], file_count="multiple" ) load_doc_btn = gr.Button("📥 加载文档", variant="secondary") mode_toggle = gr.Checkbox( label="📄 启用文档问答模式", value=False, info="开启后基于上传的文档回答问题" ) status_box = gr.Textbox( label="状态", value="就绪", interactive=False, lines=1 ) clear_btn = gr.Button("🗑️ 清除状态", variant="secondary") with gr.Column(scale=3): gr.Markdown("### 💬 对话") chatbot = gr.Chatbot( label="对话记录", height=450, placeholder="输入问题开始对话...", show_copy_button=True ) msg_input = gr.Textbox( label="输入消息", placeholder="输入您的问题,按 Enter 发送...", lines=2 ) with gr.Row(): submit_btn = gr.Button("🚀 发送", variant="primary", scale=2) clear_chat_btn = gr.Button("🧹 清空对话", variant="secondary", scale=1) gr.Markdown(""" --- **提示:** - 首次使用请在 Space 设置中配置 `DASHSCOPE_API_KEY` 环境变量 - 文档问答模式需要先上传文档并点击"加载文档" - 通义千问模型:qwen-turbo(快速)/ qwen-plus(平衡)/ qwen-max(最强) """) # ============== 事件处理 ============== def load_docs(files): """加载文档到向量数据库""" global vector_store, docs_loaded if not files: return "⚠️ 请先选择文件", None try: file_paths = [f.name for f in files] documents = load_documents(file_paths) if not documents: return "❌ 未找到可加载的文档", None vector_store = create_vector_store(documents) docs_loaded = True return f"✅ 已加载 {len(documents)} 个文档片段", vector_store except Exception as e: return f"❌ 加载失败:{str(e)}", None def respond(message, chat_history, files, use_doc_mode, current_store): """处理用户消息""" global vector_store, docs_loaded if not message: return "", chat_history, "⚠️ 请输入内容" # 检查 API 密钥 if not DASHSCOPE_API_KEY: error_msg = "⚠️ 请在 Space 设置中配置 DASHSCOPE_API_KEY 环境变量\n\n配置方法:\n1. 进入 Space Settings\n2. 找到 Repository secrets\n3. 添加 DASHSCOPE_API_KEY" if chat_history is None: chat_history = [] chat_history.append((message, error_msg)) return "", chat_history, "❌ 缺少 API 密钥" # 如果是文档模式但没有加载文档 if use_doc_mode and not docs_loaded: hint_msg = "📄 检测到您启用了文档问答模式,但尚未加载文档。\n\n请先上传文档并点击「加载文档」按钮,或者关闭文档问答模式进行普通聊天。" if chat_history is None: chat_history = [] chat_history.append((message, hint_msg)) return "", chat_history, "⚠️ 请先加载文档" # 文档问答模式 if use_doc_mode and docs_loaded and vector_store: context = retrieve_context(message) if context: prompt = f"""请根据以下文档内容回答问题。如果文档中没有相关信息,请说明。 文档内容: {context} 用户问题:{message} 请基于文档内容给出准确回答:""" messages = [ {"role": "system", "content": "你是一个专业的文档问答助手,请基于提供的文档内容准确回答问题。"}, {"role": "user", "content": prompt} ] response = chat_with_qwen(messages) if chat_history is None: chat_history = [] chat_history.append((message, response)) return "", chat_history, "✅ 基于文档回答" else: hint_msg = "🤔 文档中没有找到相关信息。\n\n您可以:\n1. 尝试换一种问法\n2. 关闭文档问答模式进行普通聊天" if chat_history is None: chat_history = [] chat_history.append((message, hint_msg)) return "", chat_history, "⚠️ 未找到相关信息" # 普通聊天模式 messages = [{"role": "system", "content": "你是一个友好、专业的 AI 助手,由通义千问驱动。"}] if chat_history: for user_msg, assistant_msg in chat_history[-5:]: messages.append({"role": "user", "content": user_msg}) messages.append({"role": "assistant", "content": assistant_msg}) messages.append({"role": "user", "content": message}) response = chat_with_qwen(messages) if chat_history is None: chat_history = [] chat_history.append((message, response)) return "", chat_history, "✅ 已回复" def clear_state(): """清除状态""" global vector_store, docs_loaded vector_store = None docs_loaded = False return "状态已清除", None, [] def clear_chat_only(): """只清空对话""" return "", [] # 绑定事件 load_doc_btn.click( fn=load_docs, inputs=[file_upload], outputs=[status_box, doc_store] ) submit_btn.click( fn=respond, inputs=[msg_input, chatbot, file_upload, mode_toggle, doc_store], outputs=[msg_input, chatbot, status_box] ) msg_input.submit( fn=respond, inputs=[msg_input, chatbot, file_upload, mode_toggle, doc_store], outputs=[msg_input, chatbot, status_box] ) clear_btn.click( fn=clear_state, outputs=[status_box, doc_store, chatbot] ) clear_chat_btn.click( fn=clear_chat_only, outputs=[msg_input, chatbot] ) if __name__ == "__main__": demo.launch()