通义千问集成

Java突击队大约 5 分钟

千问（Qwen）是阿里巴巴达摩院开发的大语言模型系列，具有强大的自然语言处理能力。Spring AI提供了对千问模型的完整支持。

3.1.1 千问模型特点

多语言支持：优秀的中文和英文处理能力
知识丰富：涵盖广泛的知识领域
性能优越：高效的推理能力
API友好：易于集成和使用

3.1.2 可用的千问模型

千问提供了多个模型版本：

qwen-turbo：高速版本，适合实时对话
qwen-plus：增强版本，知识更丰富
qwen-max：高性能版本，适合复杂任务
qwen-vl：视觉语言模型，支持图像处理

千问模型家族

3.2 配置千问客户端

3.2.1 添加依赖

在pom.xml中添加千问依赖：

<dependency>
    <groupId>org.springframework.ai</groupId>
    <artifactId>spring-ai-qwen-spring-boot-starter</artifactId>
</dependency>

3.2.2 配置API密钥

在application.properties中配置千问API密钥：

# 千问配置
spring.ai.qwen.api-key=your-qwen-api-key
spring.ai.qwen.chat.options.model=qwen-turbo
spring.ai.qwen.chat.options.temperature=0.7
spring.ai.qwen.chat.options.max-tokens=1000

3.2.3 编程配置

@Configuration
public class QwenConfig {

    @Bean
    public ChatClient qwenChatClient(ChatClient.Builder builder) {
        return builder
                .defaultSystem("You are a helpful assistant")
                .defaultOptions(QwenChatOptions.builder()
                        .withModel("qwen-plus")
                        .withTemperature(0.8f)
                        .withMaxTokens(1500)
                        .build())
                .build();
    }
}

3.3 基础对话示例

3.3.1 简单对话

@RestController
public class QwenController {

    private final ChatClient qwenChatClient;

    public QwenController(ChatClient qwenChatClient) {
        this.qwenChatClient = qwenChatClient;
    }

    @GetMapping("/qwen/chat")
    public String chat(@RequestParam String message) {
        return qwenChatClient.prompt()
                .user(message)
                .call()
                .content();
    }
}

3.3.2 系统提示对话

@GetMapping("/qwen/system-chat")
public String systemChat(@RequestParam String message) {
    return qwenChatClient.prompt()
            .system("You are a professional Spring developer")
            .user(message)
            .call()
            .content();
}

3.3.3 多轮对话

@GetMapping("/qwen/multi-turn")
public String multiTurnChat(
        @RequestParam String initialMessage,
        @RequestParam String followUpMessage) {

    ChatResponse firstResponse = qwenChatClient.prompt()
            .user(initialMessage)
            .call();

    String assistantResponse = firstResponse.getResult().getOutput().getContent();

    ChatResponse secondResponse = qwenChatClient.prompt()
            .user(followUpMessage)
            .call();

    return secondResponse.getContent();
}

3.4 高级功能

3.4.1 流式响应

千问支持流式响应，适合实时交互场景：

@GetMapping("/qwen/stream")
public Flux<ServerSentEvent<String>> streamChat(@RequestParam String message) {
    return qwenChatClient.prompt()
            .user(message)
            .stream()
            .flatMapIterable(ChatResponse::getResults)
            .map(result -> result.getOutput().getContent())
            .map(content -> ServerSentEvent.builder(content).build());
}

前端使用示例：

<!DOCTYPE html>
<html>
<head>
    <title>千问流式对话</title>
</head>
<body>
    <div id="response"></div>
    <script>
        const eventSource = new EventSource('/qwen/stream?message=Tell me about Spring AI');
        eventSource.onmessage = function(event) {
            document.getElementById('response').innerHTML += event.data + ' ';
        };
    </script>
</body>
</html>

3.4.2 函数调用

千问支持函数调用，可以集成外部工具：

@RestController
public class QwenFunctionController {

    private final ChatClient qwenChatClient;

    public QwenFunctionController(ChatClient qwenChatClient) {
        this.qwenChatClient = qwenChatClient;
    }

    @GetMapping("/qwen/function")
    public String functionCall(@RequestParam String question) {
        FunctionCallback functionCallback = new FunctionCallback(
            "getWeather",
            "Get current weather for a location",
            new Schema(SchemaType.OBJECT)
                .addProperty("location", SchemaProperty.of(SchemaType.STRING, "City name"))
                .addRequired("location"),
            this::getWeather
        );

        return qwenChatClient.prompt()
                .user(question)
                .functions(functionCallback)
                .call()
                .content();
    }

    private String getWeather(Map<String, Object> arguments) {
        String location = (String) arguments.get("location");
        // 实际天气API调用
        return "The current weather in " + location + " is sunny with 25°C";
    }
}

3.4.3 图像生成

千问视觉模型支持图像生成：

@RestController
public class QwenImageController {

    private final ImageClient imageClient;

    public QwenImageController(ImageClient imageClient) {
        this.imageClient = imageClient;
    }

    @GetMapping("/qwen/image")
    public String generateImage(@RequestParam String prompt) {
        ImagePrompt imagePrompt = new ImagePrompt(prompt);
        ImageResponse response = imageClient.generate(imagePrompt);

        return response.getResult().getOutput().getUrl();
    }
}

3.5 实际应用示例

3.5.1 智能客服系统

@Service
public class QwenCustomerService {

    private final ChatClient qwenChatClient;

    public QwenCustomerService(ChatClient qwenChatClient) {
        this.qwenChatClient = qwenChatClient;
    }

    public String handleCustomerQuery(String customerMessage) {
        return qwenChatClient.prompt()
                .system("You are a customer service representative for an e-commerce platform")
                .user(customerMessage)
                .call()
                .content();
    }
}

3.5.2 内容生成器

@RestController
public class ContentGenerator {

    private final ChatClient qwenChatClient;

    public ContentGenerator(ChatClient qwenChatClient) {
        this.qwenChatClient = qwenChatClient;
    }

    @GetMapping("/qwen/content")
    public String generateContent(
            @RequestParam String topic,
            @RequestParam String format,
            @RequestParam int length) {

        PromptTemplate template = new PromptTemplate(
            "Generate {format} content about {topic} with approximately {length} words"
        );

        Prompt prompt = template.create(
            Map.of("topic", topic, "format", format, "length", length)
        );

        return qwenChatClient.prompt(prompt).call().content();
    }
}

3.6 性能优化

3.6.1 缓存策略

@Configuration
@EnableCaching
public class QwenCacheConfig {

    @Bean
    public CacheManager cacheManager() {
        SimpleCacheManager cacheManager = new SimpleCacheManager();
        cacheManager.setCaches(Arrays.asList(
            new ConcurrentMapCache("qwen-responses")
        ));
        return cacheManager;
    }
}

3.6.2 异步调用

@RestController
public class AsyncQwenController {

    private final ChatClient qwenChatClient;

    public AsyncQwenController(ChatClient qwenChatClient) {
        this.qwenChatClient = qwenChatClient;
    }

    @GetMapping("/qwen/async")
    public CompletableFuture<String> asyncChat(@RequestParam String message) {
        return qwenChatClient.prompt()
                .user(message)
                .callAsync()
                .thenApply(ChatResponse::getContent);
    }
}

3.7 错误处理

3.7.1 异常处理

@RestControllerAdvice
public class QwenExceptionHandler {

    @ExceptionHandler(QwenApiException.class)
    public ResponseEntity<ErrorResponse> handleQwenError(QwenApiException ex) {
        ErrorResponse error = new ErrorResponse(
            "QWEN_API_ERROR",
            "千问API调用失败: " + ex.getMessage()
        );
        return new ResponseEntity<>(error, HttpStatus.SERVICE_UNAVAILABLE);
    }
}

3.7.2 重试机制

@Configuration
public class QwenRetryConfig {

    @Bean
    public ChatClient retryChatClient(ChatClient chatClient) {
        return new ChatClient() {
            @Override
            public ChatResponse call(Prompt prompt) {
                int maxAttempts = 3;
                for (int attempt = 1; attempt <= maxAttempts; attempt++) {
                    try {
                        return chatClient.call(prompt);
                    } catch (Exception ex) {
                        if (attempt == maxAttempts) {
                            throw ex;
                        }
                        try {
                            Thread.sleep(1000 * attempt);
                        } catch (InterruptedException e) {
                            Thread.currentThread().interrupt();
                        }
                    }
                }
                throw new RuntimeException("Max retry attempts reached");
            }

            // 实现其他方法...
        };
    }
}

3.8 监控与日志

3.8.1 日志记录

@Configuration
public class QwenLogging {

    @Bean
    public ChatClient loggingChatClient(ChatClient chatClient) {
        return new ChatClient() {
            @Override
            public ChatResponse call(Prompt prompt) {
                long startTime = System.currentTimeMillis();
                try {
                    ChatResponse response = chatClient.call(prompt);
                    long duration = System.currentTimeMillis() - startTime;
                    log.info("千问调用耗时: {}ms, Prompt: {}", duration, prompt);
                    return response;
                } catch (Exception ex) {
                    log.error("千问调用失败", ex);
                    throw ex;
                }
            }

            // 实现其他方法...
        };
    }
}

3.8.2 指标监控

@RestController
public class QwenMetricsController {

    private final MeterRegistry meterRegistry;

    public QwenMetricsController(MeterRegistry meterRegistry) {
        this.meterRegistry = meterRegistry;
    }

    @GetMapping("/qwen/metrics")
    public Map<String, Object> getMetrics() {
        return Map.of(
            "totalCalls", meterRegistry.counter("qwen.total.calls").count(),
            "averageDuration", meterRegistry.timer("qwen.duration").mean(),
            "errorRate", meterRegistry.counter("qwen.errors").count() /
                       (meterRegistry.counter("qwen.total.calls").count() + 1.0)
        );
    }
}

3.9 小结

本章详细介绍了千问模型的集成和使用：

模型特点：了解千问的核心优势
客户端配置：掌握千问客户端的配置方法
基础对话：学习基本的对话交互
高级功能：掌握流式响应、函数调用等高级特性
实际应用：了解千问在实际项目中的应用场景
性能优化：学习缓存、异步等优化技术
错误处理：掌握异常处理和重试机制
监控日志：了解监控和日志记录

在下一章中，我们将学习如何集成DeepSeek模型。点击这里👇🏻获取：100万QPS短链系统、复杂的商城微服务系统、智能翻译助手AI Agent、SaaS点餐系统、刷题吧小程序、商城系统、秒杀系统、AI项目、代码生成神器、苏三demo项目、智能天气播报AI Agent、智能代码审查AI Agent等 10 个项目的：项目源代码、开发教程和技术答疑