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
-
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(); } } -
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:
- ChatModel实现流式输出
- 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
-
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>-
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}
-
-
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(); } }-
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(); } }-
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
-
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>-
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}
-
-
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(); } }-
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
-
新建一个module,pom、配置文件、配置类和上面代码相同。
-
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}。 -
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
-
新建一个module,pom、配置文件、配置类和上面相同。
-
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) { } -
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); } }
本文来自博客园,作者:NE_STOP,转载请注明原文链接:https://www.cnblogs.com/alineverstop/p/23390699
浙公网安备 33010602011771号