用【Java】调用百度千帆大模型并提供流式接口【SSE】响应_java 怎么实现调用千帆模型流式输出数据

代码参考：

千帆API流式调用：PHP、JS、Nodejs、Python、Java、C# 、Go流式示例代码 - 百度智能云千帆社区本文旨在提供一个全面的指南，涵盖了在PHP、JS、Nodejs、Python、Java、C# 中流式调用千帆API的关键技巧和最佳实践。不论您是初学者还是有经验的开发者，相信这里的内容都能帮助您提升https://cloud.baidu.com/qianfandev/topic/268202

效果演示:

API服务:

接口地址:

V1版本:https://apis.ydxiaoshuai.cn/xai/rest/llm/baidu/qianfan/chat?accessToken=ACCESSTOKEN&prompt=PROMPT

V2版本使用WebFlux:https://apis.ydxiaoshuai.cn/xai/rest/llm/baidu/qianfan/v2/chat?accessToken=ACCESSTOKEN&prompt=PROMPT

请求方式:GET

参数替换:

替换ACCESSTOKEN为自己的

替换PROMPT为自己要提问的问题

注意:服务器带宽有限，请不要恶意攻击

Java-Controller代码:

/**
     * 发送问题
     *
     * @param apiKey - 用于获取token的apiKey
     * @param secretKey - 用于获取token的secretKey
     * @param accessToken - 接口token
     * @param prompt - 用于权限验证，从服务接口认证信息中获取
     * @return 百度千帆的回答
     */
    @GetMapping(value = "/baidu/qianfan/chat")
    public void baiduQianfanChat(@RequestParam(value ="apiKey",required = false) String apiKey,
                                 @RequestParam(value ="secretKey",required = false) String secretKey,
                                 @RequestParam(value ="accessToken",required = false) String accessToken,
                                 @RequestParam(value ="prompt",required = false) String prompt,
                                 HttpServletResponse res) throws Exception {
        LiteLLMResult bean = new LiteLLMResult();
        QianFanResponseDTO responseDTO = new QianFanResponseDTO();
        log.info("【百度千帆-prompt内容】：{}", prompt);
        // 响应流
        res.setHeader(HttpHeaders.CONTENT_TYPE, MediaType.TEXT_EVENT_STREAM_VALUE);
        res.setCharacterEncoding("UTF-8");
        res.setHeader(HttpHeaders.PRAGMA, "no-cache");
        ServletOutputStream out = null;
        String REQ_URL = null;
        try {
            out = res.getOutputStream();
            if (StrUtil.isEmpty(prompt)) {
                bean.fail("无效问题，请重新输入",500);
                out.write(JSON.toJSONString(bean).getBytes());
                return;
            }
            okhttp3.MediaType mediaType = okhttp3.MediaType.parse(MediaType.APPLICATION_JSON_VALUE);
            QianFanChatBean requestBean =  QianFanUtil.getRequestData(prompt);
            requestBean.setStream(true);
            RequestBody body = RequestBody.create(mediaType, JSON.toJSONString(requestBean));
            boolean flag = StrUtil.isEmpty(apiKey)||StrUtil.isEmpty(secretKey);
            if(flag && StrUtil.isNotEmpty(accessToken)){
                 REQ_URL = QianFanUtil.API_URL.replace("ACCESS_TOKEN",accessToken);
            }else if(StrUtil.isNotEmpty(apiKey)&&StrUtil.isNotEmpty(secretKey)){
                 REQ_URL = QianFanUtil.API_URL.replace("ACCESS_TOKEN",QianFanUtil.getAccessToken(apiKey,secretKey));
            }else{
                bean.fail("apiKey|secretKey|accessToken参数为空,请检查",500);
                out.write(JSON.toJSONString(bean).getBytes());
                return;
            }
            Request request = new Request.Builder()
                    .url(REQ_URL)
                    .method(Method.POST.toString(), body)
                    .addHeader(Header.CONTENT_TYPE.getValue(), MediaType.APPLICATION_JSON_VALUE)
                    .build();
            Response response = QianFanUtil.HTTP_CLIENT.newCall(request).execute();
            try (ResponseBody responseBody = response.body()) {
                if (responseBody != null) {
                    try (BufferedReader reader = new BufferedReader(responseBody.charStream())) {
                        String line;
                        while ((line = reader.readLine()) != null) {
                            String result = line.replace("data: ", "");
                            if(StrUtil.isNotEmpty(result)){
                                QianFanResponseBean qianFanResponseBean = JSON.parseObject(result,QianFanResponseBean.class);
                                responseDTO.setContent(qianFanResponseBean.getResult());
                                bean.success("执行成功",responseDTO);
                                out.write(JSON.toJSONString(bean).getBytes());
                                out.flush();
                                //防止返回内容重复或错误。暂停一下
                                Thread.sleep(100);
                            }
                        }
                    }finally {
                        responseBody.close();
                        response.close();
                        QianFanUtil.HTTP_CLIENT.connectionPool().evictAll();
                        QianFanUtil.HTTP_CLIENT.dispatcher().executorService().shutdown();
                        QianFanUtil.HTTP_CLIENT.newCall(request).cancel();
                    }
                }
            }
        } catch (Exception e) {
            bean.fail("系统内部错误，请联系管理员",500);
            out.write(JSON.toJSONString(bean).getBytes());
            return;
        } finally {
            try {
                if (out != null) {
                    out.close();
                }
            } catch (IOException e) {
                bean.fail("系统内部错误，请联系管理员",500);
                out.write(JSON.toJSONString(bean).getBytes());
                return;
            }
        }
    }

LiteLLMResult

@Data
@JsonInclude(JsonInclude.Include.NON_NULL)
public class LiteLLMResult extends BaseResponseBean {
    /**
     * 具体参数
     **/
    private XingHuoResponseDTO data;
 
    public LiteLLMResult success(String message, XingHuoResponseDTO data) {
        this.message = message;
        this.code = CommonConstant.SC_OK_200;
        this.data = data;
        return this;
    }
    public LiteLLMResult fail(String message, Integer code) {
        this.message = message;
        this.code = code;
        return this;
    }
    public LiteLLMResult error(String message) {
        this.message = message;
        this.code = CommonConstant.SC_INTERNAL_SERVER_ERROR_500;
        return this;
    }
}

BaseResponseBean

@Data
@JsonInclude(JsonInclude.Include.NON_NULL)
public class BaseResponseBean implements Serializable {
    private static final long serialVersionUID = 1L;
    private static final long timestamps = System.currentTimeMillis();
    /**
     * 返回处理消息
     */
    public String log_id = timestamps+"-"+IdUtil.fastSimpleUUID();
    /**
     * 返回处理消息
     */
    public String message = "ok";
    /**
     * 返回处理消息
     */
    public String message_zh = "操作成功！";
    /**
     * 返回代码
     */
    public Integer code = 200;
 
    /**
     * 时间戳
     */
    public long timestamp = timestamps;
 
    /**
     * 作者
     */
    public String author = "小帅丶";
 
}

QianFanResponseDTO

@Data
public class QianFanResponseDTO{
    private String uId;
    private String content;
    private String chatId;
}

Nginx配置SSE

在自己域名反向代理的location下面配置即可

		 proxy_buffering off;
		 proxy_cache off;
		 proxy_set_header Connection '';
		 chunked_transfer_encoding off;  # 开启分块传输编码
		 tcp_nopush on;  # 开启TCP NOPUSH选项，禁止Nagle算法
		 tcp_nodelay on;  # 开启TCP NODELAY选项，禁止延迟ACK算法
		 # SSE事件流
         add_header Content-Type text/event-stream;
         add_header Cache-Control no-cache;