Spring AI Alibaba -- 流式输出与Prompt提示词

ChatClient

chatClient 是 Spring AI(含 Spring AI Alibaba)提供的高阶、流式 Fluent 对话客户端,用来和大模型交互,类似 WebClient,是官方推荐的上层入口,比直接用 ChatModel 更方便。基于chatModel构建。

  • ChatModel:底层原子 API,只负责发 Prompt、拿 ChatResponse,所有逻辑自己拼,灵活、样板代码多
  • ChatClient:上层封装,把 system prompt、记忆、工具、拦截器统一串起来,适合业务开发

chatModel 示例

@RestController
public class ChatModelController {

    @Resource
    private ChatModel dashscopeChatModel;

    /**
     * 普通调用
     * @param msg
     * @return
     */
    @GetMapping("/chatmodel/dochat")
    public String dochat(@RequestParam(name="msg",defaultValue = "你是谁") String msg) {
        return dashscopeChatModel.call(msg);
    }
}

ChatClient示例

  1. 配置类

    @Configuration
    public class SaaLLMConfig {
    
        // 方式一
        @Value("${spring.ai.dashscope.api-key}")
        private String apiKey;
    
        // 配置大模型对外暴露的api
        @Bean
        public DashScopeApi getDashScopeApi() {
            return DashScopeApi.builder().apiKey(apiKey).build();
        }
    
        //chatclient不支持自动注入,只能手动注入。在配置类里手动注入。
        @Bean
        public ChatClient chatClient(ChatModel chatModel) {
          return ChatClient.builder(chatModel).build();
        }
    }
    
  2. 控制类

    @RestController
    public class ChatClientControllerV2 {
    
        @Resource
        private ChatClient chatClient;
        /**
         * 普通调用
         * @param msg
         * @return
         */
        @GetMapping("/chatclient/dochat2")
        public String dochat(@RequestParam(name="msg",defaultValue = "你是谁") String msg) {
            String content = chatClient.prompt().user(msg).call().content();
            return content;
        }
    }
    

流式输出(StreamingOutput)

一种逐步返回大模型生成结果的技术,生成一点返回一点,允许服务器将响应内容分批次实时传输给客户端,而不是等待全部内容生成完毕后再一次性返回。

这种机制能显著提升用户体验,尤其适用于大模型响应较慢的场景(比如长文本或复杂推理结果 )

Spring AI Alibaba的流式输出有两种:

  1. ChatModel实现流式输出
  2. ChatClient实现流式输出

Server-Sent Events(SSE)

Server-Sent Events 是一种允许服务端可以持续推送数据片段到前端的web技术。通过单向HTTP长连接,让服务器主动将数据推给客户端,SSE是轻量级的单向通信协议,适合AI对话这类服务端主导的场景。

AI 打字机效果很多就是用 SSE 实现的。

多种大模型在系统中共存的示例

  1. 新建一个module,pom为:

    <?xml version="1.0" encoding="UTF-8"?>
    <project xmlns="http://maven.apache.org/POM/4.0.0"
             xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
             xsi:schemaLocation="http://maven.apache.org/POM/4.0.0 http://maven.apache.org/xsd/maven-4.0.0.xsd">
        <modelVersion>4.0.0</modelVersion>
        <parent>
            <groupId>com.ali</groupId>
            <artifactId>springAiAlibaba-workV1</artifactId>
            <version>1.0-SNAPSHOT</version>
        </parent>
    
        <artifactId>ssa04-streamingOutput</artifactId>
    
        <properties>
            <maven.compiler.source>17</maven.compiler.source>
            <maven.compiler.target>17</maven.compiler.target>
            <project.build.sourceEncoding>UTF-8</project.build.sourceEncoding>
        </properties>
    
        <dependencies>
            <dependency>
                <groupId>org.springframework.boot</groupId>
                <artifactId>spring-boot-starter-web</artifactId>
            </dependency>
            <dependency>
                <groupId>com.alibaba.cloud.ai</groupId>
                <artifactId>spring-ai-alibaba-starter-dashscope</artifactId>
            </dependency>
            <dependency>
                <groupId>org.projectlombok</groupId>
                <artifactId>lombok</artifactId>
                <optional>true</optional>
            </dependency>
        </dependencies>
    </project>
    
    1. 配置文件

      server.port=8004
      # ?????????
      server.servlet.encoding.charset=UTF-8
      server.servlet.encoding.enabled=true
      server.servlet.encoding.force=true
      
      spring.application.name=ssa04-streamingOutput
      
      #springai ????config
      spring.ai.dashscope.api-key=${aliQwen-apikey}
      
  2. 配置类:

    @Configuration
    public class SaaLLMConfig {
    
        private final String DEEPSEEk_MODEL="deepseek-v3";
        private final String QWEN_MODEL="qwen-max";
    
        // 方式一
        @Value("${spring.ai.dashscope.api-key}")
        private String apiKey;
    
        // 配置大模型对外暴露的api
        @Bean(name = "deepseekChatModel")
        public ChatModel deepseekChatModel() {
        return DashScopeChatModel.builder()
                // 配置apikey
                .dashScopeApi(DashScopeApi.builder().apiKey(apiKey).build())
                // 配置模型
                .defaultOptions(DashScopeChatOptions.builder().withModel(DEEPSEEk_MODEL).build())
                .build();
        }
    
        @Bean(name = "qwenChatModel")
        public ChatModel qwenChatModel() {
            return DashScopeChatModel.builder()
                    // 配置apikey
                    .dashScopeApi(DashScopeApi.builder().apiKey(apiKey).build())
                    // 配置模型
                    .defaultOptions(DashScopeChatOptions.builder().withModel(QWEN_MODEL).build())
                    .build();
        }
    
        @Bean(name = "deepseekChatClient")
        public ChatClient deepseekChatClient(@Qualifier("deepseekChatModel") ChatModel deepseekChatModel) {
            return ChatClient.builder(deepseekChatModel)
                    .build();
        }
    
        @Bean("qwenChatClient")
        public ChatClient qwenChatClient(@Qualifier("qwenChatModel") ChatModel qwenChatModel) {
            return ChatClient.builder(qwenChatModel)
                    .build();
        }
    }
    
    1. 控制类

      @RestController
      public class StreamOutputController {
      
          @Resource(name = "deepseekChatModel")
          private ChatModel deepseekChatModel;
      
          @Resource(name = "qwenChatModel")
          private ChatModel qwenChatModel;
      
          /**
           * 普通调用
           * @param msg
           * @return
           */
          @GetMapping("/stream/chatflux1")
          public Flux<String> chatflux1(@RequestParam(name="msg",defaultValue = "你是谁") String msg) {
              return deepseekChatModel.stream(msg);
          }
      
          /**
           * 普通调用
           * @param msg
           * @return
           */
          @GetMapping("/stream/chatflux2")
          public Flux<String> chatflux2(@RequestParam(name="msg",defaultValue = "你是谁") String msg) {
              return qwenChatModel.stream(msg);
          }
      
          @Resource(name = "deepseekChatClient")
          private ChatClient deepseekChatClient;
      
          @Resource(name = "qwenChatClient")
          private ChatClient qwenChatClient;
      
          @GetMapping("/stream/chatflux3")
          public Flux<String> chatflux3(@RequestParam(name="msg",defaultValue = "你是谁") String msg) {
              return deepseekChatClient.prompt(msg).stream().content();
          }
      
          @GetMapping("/stream/chatflux4")
          public Flux<String> chatflux4(@RequestParam(name="msg",defaultValue = "你是谁") String msg) {
              return qwenChatClient.prompt(msg).stream().content();
          }
      }
      
      1. 前端页面:

        <!DOCTYPE html>
        <html>
        <head>
            <title>SSE流式ChatModel+ChatClient+多模型共存</title>
            <style>
                body {
                    font-family: Arial, sans-serif;
                    background-color: #f4f4f4;
                    margin: 0;
                    padding: 20px;
                }
        
                #messageInput {
                    width: 90%;
                    padding: 10px;
                    font-size: 16px;
                    border: 1px solid #ccc;
                    border-radius: 4px;
                    margin-bottom: 10px;
                }
        
                button {
                    padding: 10px 20px;
                    font-size: 16px;
                    background-color: #007bff;
                    color: white;
                    border: none;
                    border-radius: 4px;
                    cursor: pointer;
                }
        
                button:hover {
                    background-color: #0056b3;
                }
        
                #messages {
                    margin-top: 20px;
                    padding: 15px;
                    background-color: #f9f9f9;
                    border: 1px solid #ddd;
                    border-radius: 8px;
                    max-height: 300px;
                    overflow-y: auto;
                    box-shadow: 0 2px 4px rgba(0, 0, 0, 0.1);
                }
        
                #messages div {
                    padding: 8px 0;
                    border-bottom: 1px solid #eee;
                    font-size: 14px;
                    color: #333;
                }
        
                #messages div:last-child {
                    border-bottom: none;
                }
            </style>
        </head>
        
        <body>
            <textarea id="messageInput" rows="4" cols="50" placeholder="请输入你的问题..."></textarea><br>
            <button onclick="sendMsg()">发送提问</button>
            <div id="messages"></div>
            <script>
                function sendMsg()
                {
                    // 获取用户输入的消息
                    const message = document.getElementById('messageInput').value;
                    if (message == "") return false;
        
                    //1 客户端使用 JavaScript 的 EventSource 对象连接到服务器上的一个特定端点(URL)
                    const eventSource = new EventSource('stream/chatflux2?msg='+message);
                    //2 监听消息事件
                    eventSource.onmessage = function (event)
                    {
                        //2.1 获得流式返回的数据
                        const data = event.data;
                        //2.2 将收获到的数据展示到页面回答窗口上
                        const messagesDiv = document.getElementById('messages');
                        //2.3 类似java里面append追加的形式将数据批次的显示
                        messagesDiv.innerHTML += data;
                    }
        
                    //3 监听错误事件,当连接发生错误(包括断开)时触发。
                    eventSource.onerror = function (error){
                        console.error('EventSource发生了错误: ',error);
                        eventSource.close();//关闭链接
                    }
                }
            </script>
        </body>
        </html>
        

Prompt提示词

Prompt 就是你发给大模型的指令、问题、上下文,用来告诉 AI 要做什么、怎么输出。

提示词的四大角色:

USER("user"),// 用户的原始提问输入,用户向AI提出的问题、命令和陈述。
ASSISTANT("assistant"), // AI返回的响应信息、记忆对话,积累回答
SYSTEM("system"),// 设定AI大模型的行为/边界/定位,指导AI的行为和响应方式,
TOOL("tool"); // 桥接外部服务,类似调用第三方工具类,可进行函数调用,支付/数据查询等操作

代码示例

  1. 新建module,pom为:

    <?xml version="1.0" encoding="UTF-8"?>
    <project xmlns="http://maven.apache.org/POM/4.0.0"
             xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
             xsi:schemaLocation="http://maven.apache.org/POM/4.0.0 http://maven.apache.org/xsd/maven-4.0.0.xsd">
        <modelVersion>4.0.0</modelVersion>
        <parent>
            <groupId>com.ali</groupId>
            <artifactId>springAiAlibaba-workV1</artifactId>
            <version>1.0-SNAPSHOT</version>
        </parent>
    
        <artifactId>saa05-chatPrompt</artifactId>
    
        <properties>
            <maven.compiler.source>17</maven.compiler.source>
            <maven.compiler.target>17</maven.compiler.target>
            <project.build.sourceEncoding>UTF-8</project.build.sourceEncoding>
        </properties>
        <dependencies>
            <dependency>
                <groupId>org.springframework.boot</groupId>
                <artifactId>spring-boot-starter-web</artifactId>
            </dependency>
            <dependency>
                <groupId>com.alibaba.cloud.ai</groupId>
                <artifactId>spring-ai-alibaba-starter-dashscope</artifactId>
            </dependency>
            <dependency>
                <groupId>org.projectlombok</groupId>
                <artifactId>lombok</artifactId>
                <optional>true</optional>
            </dependency>
        </dependencies>
    </project>
    
    1. 配置文件

      server.port=8005
      # ?????????
      server.servlet.encoding.charset=UTF-8
      server.servlet.encoding.enabled=true
      server.servlet.encoding.force=true
      
      spring.application.name=ssa05-chatPrompt
      
      #springai ????config
      spring.ai.dashscope.api-key=${aliQwen-apikey}
      
  2. 配置类

    @Configuration
    public class SaaLLMConfig {
    
        private final String DEEPSEEk_MODEL="deepseek-v3";
        private final String QWEN_MODEL="qwen-max";
    
        // 方式一
        @Value("${spring.ai.dashscope.api-key}")
        private String apiKey;
    
        // 配置大模型对外暴露的api
        @Bean(name = "deepseekChatModel")
        public ChatModel deepseekChatModel() {
        return DashScopeChatModel.builder()
                // 配置apikey
                .dashScopeApi(DashScopeApi.builder().apiKey(apiKey).build())
                // 配置模型
                .defaultOptions(DashScopeChatOptions.builder().withModel(DEEPSEEk_MODEL).build())
                .build();
        }
    
        @Bean(name = "qwenChatModel")
        public ChatModel qwenChatModel() {
            return DashScopeChatModel.builder()
                    // 配置apikey
                    .dashScopeApi(DashScopeApi.builder().apiKey(apiKey).build())
                    // 配置模型
                    .defaultOptions(DashScopeChatOptions.builder().withModel(QWEN_MODEL).build())
                    .build();
        }
    
        @Bean(name = "deepseekChatClient")
        public ChatClient deepseekChatClient(@Qualifier("deepseekChatModel") ChatModel deepseekChatModel) {
            return ChatClient.builder(deepseekChatModel)
                    .build();
        }
    
        @Bean("qwenChatClient")
        public ChatClient qwenChatClient(@Qualifier("qwenChatModel") ChatModel qwenChatModel) {
            return ChatClient.builder(qwenChatModel)
                    .build();
        }
    }
    
    1. 控制类

      @RestController
      public class PromptController {
      
          @Resource(name = "deepseekChatModel")
          private ChatModel deepseekChatModel;
      
          @Resource(name = "qwenChatModel")
          private ChatModel qwenChatModel;
      
          @Resource(name = "deepseekChatClient")
          private ChatClient deepseekChatClient;
      
          @Resource(name = "qwenChatClient")
          private ChatClient qwenChatClient;
      
          /**
           * 普通调用
           * @param msg
           * @return
           */
          @GetMapping("/prompt/chat")
          public Flux<String> chat(@RequestParam(name="msg",defaultValue = "你是谁") String msg) {
              return deepseekChatClient.prompt()
                      // AI角色设定/能力边界
                      .system("你是一个法律助手,只回答法律问题,其他问题请咨询其他人。")
                      .user(msg)
                      .stream()
                      .content();
          }
      
          @GetMapping("/prompt/chat2")
          public Flux<ChatResponse> chat2(@RequestParam(name="msg",defaultValue = "你是谁") String msg) {
              // AI角色设定/能力边界
              SystemMessage sysexMessage = new SystemMessage("你是一个讲故事助手,每个故事控制在300字以内");
              // 用户输入
              UserMessage userMessage = new UserMessage(msg);
      
              Prompt prompt = new Prompt(sysexMessage, userMessage);
      
              return deepseekChatModel.stream(prompt);
          }
      
          @GetMapping("/prompt/chat3")
          public Flux<String> chat3(@RequestParam(name="msg",defaultValue = "你是谁") String msg) {
              // AI角色设定/能力边界
              SystemMessage sysexMessage = new SystemMessage("你是一个讲故事助手,每个故事控制在300字以内");
              // 用户输入
              UserMessage userMessage = new UserMessage(msg);
      
              Prompt prompt = new Prompt(sysexMessage, userMessage);
      
              return deepseekChatModel.stream(prompt).map(respone->respone.getResult().getOutput().getText());
          }
      
          // 返回大模型的回复
          @GetMapping("/prompt/chat4")
          public String chat4(@RequestParam(name="msg",defaultValue = "你是谁") String msg) {
              AssistantMessage assistantMessage = deepseekChatClient.prompt()
                      .user(msg)
                      .call()
                      .chatResponse()
                      .getResult()
                      .getOutput();
              return assistantMessage.getText();
          }
      
          @GetMapping("/prompt/chat5")
          public String chat5(@RequestParam(name="msg",defaultValue = "今天北京的天气如何") String msg) {
              String answer = deepseekChatClient.prompt()
                      .user(msg)
                      .call()
                      .chatResponse()
                      .getResult()
                      .getOutput()
                      .getText();
              // ToolResponseMessage 调用第三方工具
              ToolResponseMessage toolResponseMessage = new ToolResponseMessage(
                      List.of(new ToolResponseMessage.ToolResponse
                              ("1", "获得天气",msg)
                      )
              );
              String toolresponse = toolResponseMessage.getText();
              String result = "大模型回答:" + answer + "\n" + "工具调用结果:" + toolresponse;
              return result;
          }
      
      }
      

prompt template提示词模板

引入占位符以动态内容插入,比如入职邀请函模板

代码应用

  1. 新建一个module,pom、配置文件、配置类和上面代码相同。

  2. 在resources 目录下新增userTemplate.txt 模板,内容如下:

    讲一个关于{topic}的故事,输出格式为{output_format}。
    
  3. 控制类

    @RestController
    public class PromptTemplateController {
    
        @Resource(name = "deepseekChatModel")
        private ChatModel deepseekChatModel;
    
        @Resource(name = "qwenChatModel")
        private ChatModel qwenChatModel;
    
        @Resource(name = "deepseekChatClient")
        private ChatClient deepseekChatClient;
    
        @Resource(name = "qwenChatClient")
        private ChatClient qwenChatClient;
    
        /**
         * promptTemplate的基本使用
         * @param
         * @return
         */
        @GetMapping("/promptTemplate/chat")
        public Flux<String> chat(@RequestParam("topic") String topic,
                                 @RequestParam("output_format") String output_format,
                                 @RequestParam("wordCount") String wordCount) {
            PromptTemplate promptTemplate = new PromptTemplate(
                    "你是一个讲故事助手," +
                    "请根据用户提供的主题{topic}," +
                    "生成一个{wordCount}字以内的故事," +
                    "并且输出格式为{output_format}。");
            Prompt prompt = promptTemplate.create(Map.of(
                    "topic", topic,
                    "output_format", output_format,
                    "wordCount", wordCount
                    ));
            return deepseekChatClient.prompt(prompt).stream().content();
        }
    
        // resource目录下的promptTemplate下的userTemplate.txt文件
        @Value("classpath:promptTemplate/userTemplate.txt")
        private org.springframework.core.io.Resource userTemplate;
    
        // 读取模板文件使用模板功能
        @GetMapping("/promptTemplate/chat2")
        public Flux<String> chat2(@RequestParam("topic") String topic,
                                 @RequestParam("output_format") String output_format,
                                 @RequestParam("wordCount") String wordCount) {
            PromptTemplate promptTemplate = new PromptTemplate(userTemplate);
            Prompt prompt = promptTemplate.create(Map.of(
                    "topic", topic,
                    "output_format", output_format
            ));
            return deepseekChatClient.prompt(prompt).stream().content();
        }
    
    
        /**
         * 使用系统模板和用户模板
         * @param sysTopic
         * @param userTopic
         * @return
         */
        @GetMapping("/promptTemplate/chat3")
        public String chat3(@RequestParam(name="sysTopic") String sysTopic,
                                  @RequestParam(name="userTopic") String userTopic) {
            // 使用系统模板和用户模板
            SystemPromptTemplate systemPromptTemplate =
                    new SystemPromptTemplate("你是一个{systemTopic}助手,只回答{systemTopic}相关问题,其他无可奉告,以HTML格式输出");
            Message systemMessage = systemPromptTemplate.createMessage(Map.of("systemTopic", sysTopic));
            PromptTemplate userPromptTemplate = new PromptTemplate("解释一下{userTopic}");
            Message userMessage = userPromptTemplate.createMessage(Map.of("userTopic", userTopic));
            // 将系统消息和用户消息组合成一个Prompt对象
            Prompt prompt = new Prompt(List.of(systemMessage, userMessage));
            return qwenChatClient.prompt(prompt).call().content();
        }
    
        // 返回大模型的回复
        @GetMapping("/promptTemplate/chat4")
        public String chat4(@RequestParam(name="msg",defaultValue = "你是谁") String msg) {
            AssistantMessage assistantMessage = deepseekChatClient.prompt()
                    .user(msg)
                    .call()
                    .chatResponse()
                    .getResult()
                    .getOutput();
            return assistantMessage.getText();
        }
    
        @GetMapping("/prompt/chat5")
        public String chat5(@RequestParam(name="msg",defaultValue = "今天北京的天气如何") String msg) {
            String answer = deepseekChatClient.prompt()
                    .user(msg)
                    .call()
                    .chatResponse()
                    .getResult()
                    .getOutput()
                    .getText();
            // ToolResponseMessage 调用第三方工具
            ToolResponseMessage toolResponseMessage = new ToolResponseMessage(
                    List.of(new ToolResponseMessage.ToolResponse
                            ("1", "获得天气",msg)
                    )
            );
            String toolresponse = toolResponseMessage.getText();
            String result = "大模型回答:" + answer + "\n" + "工具调用结果:" + toolresponse;
            return result;
        }
    
    }
    

格式化输出

让大模型不再自由写自然文字,严格按照你指定的结构输出内容(JSON、XML、java类、Markdown 表格等),最常用是输出 JSON。

代码示例

  1. 新建一个module,pom、配置文件、配置类和上面相同。

  2. 新建两个实体类Book、StudentRecord

    @Data
    public class Book {
    
        int id;
        String bookName;
    
        public Book(int id, String bookName) {
            this.id = id;
            this.bookName = bookName;
        }
    
        public int getId() {
            return id;
        }
    
        public void setId(int id) {
            this.id = id;
        }
    
        public String getBookName() {
            return bookName;
        }
    
        public void setBookName(String bookName) {
            this.bookName = bookName;
        }
    
        @Override
        public int hashCode() {
            return Objects.hash(id, bookName);
        }
    
        @Override
        public String toString() {
              return "Book{" +
                    "id=" + id +
                    ", bookName='" + bookName + '\'' +
                    '}';
        }
    }
    
    public record StudentRecord(String id, String sname, String major,String email) {
    }
    
  3. 控制类

    @RestController
    public class StructerOutputController {
    
        @Resource(name = "deepseekChatModel")
        private ChatModel deepseekChatModel;
    
        @Resource(name = "qwenChatModel")
        private ChatModel qwenChatModel;
    
        @Resource(name = "deepseekChatClient")
        private ChatClient deepseekChatClient;
    
        @Resource(name = "qwenChatClient")
        private ChatClient qwenChatClient;
    
        /**
         * 格式化输出
         * @param
         * @return
         */
        @GetMapping("/structerOutput/chat")
        public StudentRecord chat(@RequestParam("sname") String sname,
                                  @RequestParam("email") String email) {
    
            return qwenChatClient.prompt().user(new Consumer<ChatClient.PromptUserSpec>() {
                @Override
                public void accept(ChatClient.PromptUserSpec promptUserSpec) {
                    promptUserSpec.text("学号001,姓名{sname},大学计算机科学与技术,邮箱{email}")
                            .param("sname", sname)
                            .param("email", email);
                }
            }).call().entity(StudentRecord.class);
        }
    
    
        /**
         *
         * @param
         * @return
         */
        @GetMapping("/structerOutput/chat2")
        public StudentRecord chat2(@RequestParam("sname") String sname,
                                  @RequestParam("email") String email) {
    
            return qwenChatClient.prompt().user(
                    promptUserSpec -> promptUserSpec.text("学号002,姓名{sname},大学软件工程,邮箱{email}")
                    .param("sname", sname)
                    .param("email", email)).call().entity(StudentRecord.class);
        }
    
    
    }
    
posted @ 2026-10-21 18:03  NE_STOP  阅读(40)  评论(0)    收藏  举报