关于Java调用大模型方式的讨论，欢迎各位大佬留言指导

最近在搞大模型应用的开发，目前用的是Java语言。我们与大模型调用交互的方式是基于SSE（Server-Sent Events）协议。目前的方案是引用了okhttp-see的包。因为是刚刚接触，所以发出这篇文章描述一下我是如何实现的，希望业内的大佬给指点一下，我这种方式如何？有没有标准的调用代码，有没有更优雅的代码。正确的调用模型接口是怎么样的？

<dependency>
	<groupId>com.squareup.okhttp3</groupId>
	<artifactId>okhttp-sse</artifactId>
</dependency>

具体实现代码如下：

EventSourceListener 的实现：

// 写了一个基础类，定义了一些基础的变量
@Data
public abstract class BaseSSEListener extends EventSourceListener {
    private CountDownLatch countDownLatch = new CountDownLatch(1);
    // 保存会话信息
    public static ConcurrentHashMap<String, EventSource> listenerHashMap = new ConcurrentHashMap<>();
}

// 具体的实现类
@Slf4j
@Data
public class SSEPictureGeneratedSpeechListener extends BaseSSEListener {
		// 会话的id，会返给调用者，调用者可以根据这个id调用停止接口
    private String eventId;
    private HttpServletResponse response;

    public SSEPictureGeneratedSpeechListener(String eventId, HttpServletResponse response) {
        this.eventId = eventId;
        this.response = response;
    }

    /**
     * {@inheritDoc}
     * 建立sse连接
     */
    @Override
    public void onOpen(final EventSource eventSource, final Response
            response) {
        listenerHashMap.put(eventId, eventSource);
        log.info("与模型建立sse连接...");
    }

    /**
     * 事件
     *
     * @param eventSource
     * @param id
     * @param type
     * @param data
     */
    @Override
    public void onEvent(EventSource eventSource, String id, String type, String data) {
        try {
            if (response != null) {
                if ("\n".equals(data)) {
                    response.getWriter().write("id:" + eventId + "\n");
                    response.getWriter().write("event:" + "message" + "\n");
                    response.getWriter().write("data:\n\n");
                    response.getWriter().flush();
                } else {
                    response.getWriter().write("id:" + eventId + "\n");
                    response.getWriter().write("event:" + "message" + "\n");
                    response.getWriter().write("data:" + data + "\n\n");
                    response.getWriter().flush();
                }

            }
        } catch (Exception e) {
            log.error("消息错误]", e);
            getCountDownLatch().countDown();
            listenerHashMap.remove(eventId);
            throw new RuntimeException(e);
        }

    }


// 经过测试和看代码，发现不管是正常停止还是异常停止都会走onClosed接口
    @Override
    public void onClosed(final EventSource eventSource) {
        log.info("连接关闭：{}", JsonUtils.toJsonString(eventSource));
        getCountDownLatch().countDown();
        listenerHashMap.remove(eventId);
    }

// 异常情况，调用了EventSource的cancel()方法会进入这里，但是模型服务异常了也会进入这个方法。我发现response的响应码不一样，就根据这个区分的，200是用户主动停止的
    @Override
    public void onFailure(final EventSource eventSource, final Throwable t, final Response response) {
        log.error("使用事件源时出现异常...：{}", t != null ? t.getMessage() : "未知错误");
        getCountDownLatch().countDown();
        listenerHashMap.remove(eventId);
        if (response.code() == 200) {
            try {
                this.response.getWriter().write("id:" + eventId + "\n");
                this.response.getWriter().write("event:" + "cancel" + "\n");
                this.response.getWriter().write("data:" + "\n\n");
            } catch (IOException e) {
                throw new RuntimeException(e);
            }
        } else {
            try {
                this.response.getWriter().write("id:" + eventId + "\n");
                this.response.getWriter().write("event:" + "error" + "\n");
                this.response.getWriter().write("data:" + "\n\n");
            } catch (IOException e) {
                throw new RuntimeException(e);
            }
        }
    }
}

下面是controller的代码

// 停止接口
@GetMapping("stopOutput/{eventId}")
    public AjaxResult stopOutput(@PathVariable("eventId") String eventId) {
        EventSource eventSource = BaseSSEListener.listenerHashMap.get(eventId);
        if (eventSource != null) {
            eventSource.cancel();
        }
        return AjaxResult.success();
    }

// 调用的接口
@PostMapping("pictureGeneratedSpeech")
    public void pictureGeneratedSpeech(@RequestBody PictureGeneratedSpeechAO pictureGeneratedSpeechAO, HttpServletResponse response) throws IOException {
        // 开启流式返回
        response.setContentType("text/event-stream");
        response.setCharacterEncoding("UTF-8");
        response.setStatus(200);
        log.info("建立sse连接...");

        // 生成id
        Long id = IdGenerateUtil.nextId();
        HashMap<String, Object> map = new HashMap<>();

        // 下载图片
        InputStream inputStream = null;
        try {
            inputStream = FileUtils.downloadFile(storageUrl + pictureGeneratedSpeechAO.getUrl(), null);
            String base64String = Base64.getEncoder().encodeToString(FileUtils.inputStreamToByteArray(inputStream));
            map.put("imageBase64", base64String);
        } catch (Exception e) {
            e.printStackTrace();
        } finally {
            if (inputStream != null) {
                inputStream.close();
            }
        }

        map.put("prompt_text", pictureGeneratedSpeechAO.getPrompt());
        map.put("stream", true);
        try {
            SSEPictureGeneratedSpeechListener sseListener = new SSEPictureGeneratedSpeechListener(id + "", response);
            ExecuteSSEUtil.executeListener(lagerModelToolUrl + "/internvl/infer", null, sseListener, JSON.toJSONString(map));
        } catch (Exception e) {
            log.error("请求SSE错误处理", e);
            throw new BaseException("请求SSE错误处理");
        }
    }