开源模型应用落地-业务整合篇（一）

charles_vaez2024-01-19 13:14

一、前言

经过对qwen-7b-chat的部署以及与vllm的推理加速的整合，我们成功构建了一套高性能、高可靠、高安全的AI服务能力。现在，我们将着手整合具体的业务场景，以实现完整可落地的功能交付。

作为上游部门，通常会采用最常用的方式来接入下游服务。为了调用我们的AI服务，我们将使用Java语言，并分别使用HttpClient、OkHttp等工具来实现调用。这样可以确保我们能够高效地与AI服务进行交互。

二、术语

2.1.OkHttp

是一个开源的Java和Kotlin HTTP客户端库，用于进行网络请求。OkHttp支持HTTP/1.1和HTTP/2协议，具有连接池、请求重试、缓存、拦截器等功能。它还提供了异步和同步请求的支持，并且可以与各种平台和框架无缝集成，是Android开发中常用的网络请求库之一。通过使用OkHttp，开发人员可以轻松地发送HTTP请求、处理响应以及管理网络连接，从而加快应用程序的网络通信速度和效率。

2.2.HttpClient

是一个用于发送HTTP请求和接收HTTP响应的开源库。它提供了一种方便的方式来与Web服务器进行通信，并执行各种HTTP操作，例如发送GET请求、POST请求等。HttpClient库通常用于编写客户端应用程序或服务，这些应用程序需要与Web服务器或Web API进行通信。它提供了许多功能，包括连接管理、身份验证、请求和响应拦截、Cookie管理等。

2.3.HttpURLConnection

是Java提供的一个用于发送HTTP请求和接收HTTP响应的类。它是Java标准库中的一部分，用于与Web服务器进行通信。HttpURLConnection类提供了一组方法，使您能够创建HTTP连接、设置请求方法（如GET、POST等）、设置请求头、设置请求体和其他参数，并发送请求到指定的URL。它还提供了方法来获取HTTP响应的状态码、响应头和响应体等信息。

OkHttp和HttpClient提供了更丰富的功能和更好的性能，适用于大多数情况下。HttpURLConnection是Java标准库中的类，提供了基本的HTTP功能

三、前置条件

3.1. 完成Qwen-7b-Chat（Qwen-1_8B-Chat）模型的本地部署或服务端部署

参见"开源模型应用落地-qwen-7b-chat与vllm实现推理加速的正确姿势"系列文章

3.2. 完成对外服务接口的封装，屏蔽不同模型的调用差异

参见"开源模型应用落地-qwen-7b-chat与vllm实现推理加速的正确姿势"系列文章

四、技术实现

4.1. HttpURLConnection

java 复制代码 
import lombok.extern.slf4j.Slf4j;
import java.io.ByteArrayOutputStream;
import java.io.InputStream;
import java.io.OutputStream;
import java.net.HttpURLConnection;
import java.net.URL;
import java.nio.charset.StandardCharsets;
import java.util.Objects;

@Slf4j
public class QWenCallTest {

    private static String url = "http://127.0.0.1:9999/api/chat";
    private static String DEFAULT_TEMPLATE = "{\"prompt\":\"%s\",\"history\":%x,\"top_p\":0.9, \"temperature\":0.45,\"repetition_penalty\":1.1, \"max_new_tokens\":8192}";
    private static String DEFAULT_USERID = "xxxxx";
    private static String DEFAULT_SECRET = "xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx";
    private static int DEFAULT_CONNECTION_TIMEOUT = 3 * 1000;
    private static int DEFAULT_READ_TIMEOUT = 30 * 1000;

    public static void main(String[] args) {
        String question = "我家周边有什么好吃、好玩的地方嘛？";
        String history = "[{\n" +
                "\"from\": \"user\",\n" +
                "\"value\": \"你好\"\n" +
                "},\n" +
                "{\n" +
                "\"from\": \"assistant\",\n" +
                "\"value\": \"你好！有什么我能为你效劳的吗？\"\n" +
                "},\n" +
                "{\n" +
                "\"from\": \"user\",\n" +
                "\"value\": \"我家在广州，你呢?\"\n" +
                "},\n" +
                "{\n" +
                "\"from\": \"assistant\",\n" +
                "\"value\": \"我是一个人工智能助手，没有具体的家。\"\n" +
                "}]";

        String prompt = DEFAULT_TEMPLATE.replace("%s", question).replace("%x", history);
        log.info("prompt: {}", prompt);

        HttpURLConnection conn = null;
        OutputStream os = null;
        try {
            //1.设置URL
            URL urlObject = new URL(url);
            //2.打开URL连接
            conn = (HttpURLConnection) urlObject.openConnection();
            //3.设置请求方式
            conn.setRequestMethod("POST");
            conn.setRequestProperty("Content-Type", "application/json;charset=utf-8");
            conn.setRequestProperty("Accept", "text/event-stream");
            conn.setRequestProperty("userId", DEFAULT_USERID);
            conn.setRequestProperty("secret", DEFAULT_SECRET);
            conn.setDoOutput(true);
            conn.setDoInput(true);

            // 设置连接超时时间为60秒
            conn.setConnectTimeout(DEFAULT_CONNECTION_TIMEOUT);

            // 设置读取超时时间为60秒
            conn.setReadTimeout(DEFAULT_READ_TIMEOUT);


            os = conn.getOutputStream();
            os.write(prompt.getBytes("utf-8"));

        } catch (Exception e) {
            log.error("请求模型接口异常", e);
        } finally {
            if(!Objects.isNull(os)){
                try {
                    os.flush();
                    os.close();
                } catch (Exception e) {
                }

            }
        }

        InputStream is = null;
        try{
            if(!Objects.isNull(conn)){
                int responseCode = conn.getResponseCode();
                log.info("Response Code: " + responseCode);

                if(responseCode == 200){
                    is = conn.getInputStream();
                }else{
                    is = conn.getErrorStream();
                }

                byte[] bytes = new byte[1024];
                int len = 0;

                while ((len = is.read(bytes)) != -1) {
                    ByteArrayOutputStream outputStream = new ByteArrayOutputStream();
                    outputStream.write(bytes, 0, len);
                    String response = new String(outputStream.toByteArray(), StandardCharsets.UTF_8);
                    log.info(response);
                }
            }

        } catch (Exception e) {
            log.error("请求模型接口异常", e);
        } finally {
            if (!Objects.isNull(is)) {
                try {
                    is.close();
                } catch (Exception e) {
                    e.printStackTrace();
                }
            }

        }

    }

}

4.2. OkHttp

java 复制代码 
import com.alibaba.fastjson.JSON;
import lombok.extern.slf4j.Slf4j;
import okhttp3.*;
import java.io.ByteArrayOutputStream;
import java.io.InputStream;
import java.nio.charset.StandardCharsets;
import java.util.Objects;
import java.util.concurrent.CountDownLatch;
import java.util.concurrent.TimeUnit;

@Slf4j
public class QWenCallTest {
    private static String url = "http://127.0.0.1:9999/api/chat";
    private static String DEFAULT_TEMPLATE = "{\"prompt\":\"%s\",\"history\":%x,\"top_p\":0.9, \"temperature\":0.45,\"repetition_penalty\":1.2, \"max_new_tokens\":8192}";
    private static long DEFAULT_CONNECTION_TIMEOUT = 3 * 1000;
    private static long DEFAULT_WRITE_TIMEOUT = 15 * 1000;
    private static long DEFAULT_READ_TIMEOUT = 15 * 1000;
    
    private final static Request.Builder buildHeader(Request.Builder builder) {
        return builder
                .addHeader("Content-Type", "application/json; charset=utf-8")
                .addHeader("userId", "xxxxx")
                .addHeader("secret", "xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx");
    }

    private final static Request buildRequest(String prompt) {
        //创建一个请求体对象(body)
        MediaType mediaType = MediaType.parse("application/json");
        RequestBody requestBody = RequestBody.create(mediaType,prompt);

        return buildHeader(new Request.Builder().post(requestBody))
                .url(url).build();
    }

    

    public static void chat(String question,String history,CountDownLatch countDownLatch) {
        //定义请求的参数
        String prompt = DEFAULT_TEMPLATE.replace("%s", question).replace("%x", history);
        log.info("prompt: {}", prompt);

        //创建一个请求对象
        Request request = buildRequest(prompt);

        //发送请求:创建了一个请求工具对象，调用执行request对象
        OkHttpClient okHttpClient = new OkHttpClient().newBuilder()
                .connectTimeout(DEFAULT_CONNECTION_TIMEOUT, TimeUnit.MILLISECONDS)
                .writeTimeout(DEFAULT_WRITE_TIMEOUT, TimeUnit.MILLISECONDS)
                .readTimeout(DEFAULT_READ_TIMEOUT, TimeUnit.MILLISECONDS)
                .build();

        InputStream is = null;
        try {
            Response response = okHttpClient.newCall(request).execute();
            //正常返回
            if(response.code() == 200){
                //打印返回的字符数据
                is = response.body().byteStream();
                byte[] bytes = new byte[1024];
                int len = 0;
                while ((len = is.read(bytes)) != -1) {
                    ByteArrayOutputStream outputStream = new ByteArrayOutputStream();
                    outputStream.write(bytes, 0, len);
                    outputStream.flush();

                    String result = new String(outputStream.toByteArray(), StandardCharsets.UTF_8);
                    log.info(result);
                }
            }
            else{
                String result = response.body().string();
                String jsonStr = JSON.parseObject(result).toJSONString();
                log.info(jsonStr);
            }


        } catch (Throwable e) {
            log.error("执行异常", e);
        } finally {
            if (!Objects.isNull(is)) {
                try {
                    is.close();
                } catch (Exception e) {
                    e.printStackTrace();
                }
            }

            countDownLatch.countDown();

        }

    }

    public static void main(String[] args) {
        CountDownLatch countDownLatch = new CountDownLatch(1);


        String question = "我家周边有什么好吃、好玩的地方嘛？";
        String history = "[{\n" +
                "\"from\": \"user\",\n" +
                "\"value\": \"你好\"\n" +
                "},\n" +
                "{\n" +
                "\"from\": \"assistant\",\n" +
                "\"value\": \"你好！有什么我能为你效劳的吗？\"\n" +
                "},\n" +
                "{\n" +
                "\"from\": \"user\",\n" +
                "\"value\": \"我家在广州，你呢?\"\n" +
                "},\n" +
                "{\n" +
                "\"from\": \"assistant\",\n" +
                "\"value\": \"我是一个人工智能助手，没有具体的家。\"\n" +
                "}]";



        //流式输出
        long starttime = System.currentTimeMillis();
        chat(question,history,countDownLatch);
        long endtime = System.currentTimeMillis();
        System.err.println((endtime-starttime));

        try {
            countDownLatch.await();
        } catch (InterruptedException e) {
            e.printStackTrace();
        }
    }


}

maven依赖

java 复制代码 
<dependency>
    <groupId>com.squareup.okhttp3</groupId>
    <artifactId>okhttp</artifactId>
    <version>3.14.9</version>
</dependency>

4.3. HttpClient

java 复制代码 
import lombok.extern.slf4j.Slf4j;
import org.asynchttpclient.AsyncHttpClient;
import org.asynchttpclient.AsyncHttpClientConfig;
import org.asynchttpclient.DefaultAsyncHttpClient;
import org.asynchttpclient.DefaultAsyncHttpClientConfig;
import org.asynchttpclient.channel.DefaultKeepAliveStrategy;
import java.io.IOException;
import java.util.concurrent.CountDownLatch;
import java.util.concurrent.TimeUnit;

@Slf4j
public class QwenCallTest {
    private static String url = "http://127.0.0.1:9999/api/chat";
    private static String DEFAULT_TEMPLATE = "{\"prompt\":\"%s\",\"history\":%x,\"top_p\":0.9, \"temperature\":0.45,\"repetition_penalty\":1.1, \"max_new_tokens\":8192}";
    private static int DEFAULT_CONNECTION_TIMEOUT = 3 * 1000;
    private static int DEFAULT_REQUEST_TIMEOUT = 15* 1000;
    private static int DEFAULT_READ_TIMEOUT = 15* 1000;

    private static String DEFAULT_USERID = "xxxxx";
    private static String DEFAULT_SECRET = "xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx";


    private static AsyncHttpClientConfig asyncHttpClientConfig = new DefaultAsyncHttpClientConfig.Builder()
            .setConnectTimeout(DEFAULT_CONNECTION_TIMEOUT)
            .setReadTimeout(DEFAULT_READ_TIMEOUT)
            .setRequestTimeout(DEFAULT_REQUEST_TIMEOUT)
            .setTcpNoDelay(true)
            .setMaxConnections(1_000_000)
            .setMaxConnectionsPerHost(100_000)
            .setMaxRequestRetry(0)
            .setSoReuseAddress(true)
            .setKeepAlive(true)
            .setKeepAliveStrategy(new DefaultKeepAliveStrategy())
            .build();

    public static final AsyncHttpClient ugc_asyncHttpClient = new DefaultAsyncHttpClient(asyncHttpClientConfig);


    public static void chat(String question,String history, CountDownLatch countDownLatch ) {
        String prompt = DEFAULT_TEMPLATE.replace("%s", question).replace("%x", history);
        log.info("prompt: {}", prompt);

        try {
            ugc_asyncHttpClient.preparePost(url)
                    .addHeader("Content-Type", "application/json; charset=utf-8")
                    .addHeader("userId", DEFAULT_USERID)
                    .addHeader("secret", DEFAULT_SECRET)
                    .addHeader("Accept", "text/event-stream")
                    .setBody(prompt)
                    .execute(new QwenStreamHandler(countDownLatch))
                    .get(30, TimeUnit.SECONDS);

        } catch (Exception e) {
            log.error(prompt + " >> 出现异常");
        }

    }


    public static void main(String[] args) {
        CountDownLatch countDownLatch = new CountDownLatch(1);
        String question = "我家周边有什么好吃、好玩的地方嘛？";
        String history = "[{\n" +
                "\"from\": \"user\",\n" +
                "\"value\": \"你好\"\n" +
                "},\n" +
                "{\n" +
                "\"from\": \"assistant\",\n" +
                "\"value\": \"你好！有什么我能为你效劳的吗？\"\n" +
                "},\n" +
                "{\n" +
                "\"from\": \"user\",\n" +
                "\"value\": \"我家在广州，你呢?\"\n" +
                "},\n" +
                "{\n" +
                "\"from\": \"assistant\",\n" +
                "\"value\": \"我是一个人工智能助手，没有具体的家。\"\n" +
                "}]";

        chat(question,history,countDownLatch);

        try {
            countDownLatch.await();
        } catch (InterruptedException e) {
            e.printStackTrace();
        }

        try {
            ugc_asyncHttpClient.close();
        } catch (IOException e) {
            e.printStackTrace();
        }
    }


}
java 复制代码 
import io.netty.handler.codec.http.HttpHeaders;
import lombok.extern.slf4j.Slf4j;
import org.asynchttpclient.HttpResponseBodyPart;
import org.asynchttpclient.HttpResponseStatus;
import org.asynchttpclient.handler.StreamedAsyncHandler;
import org.asynchttpclient.netty.EagerResponseBodyPart;
import org.reactivestreams.Publisher;
import org.reactivestreams.Subscriber;
import org.reactivestreams.Subscription;

import java.util.concurrent.CountDownLatch;

@Slf4j
public class QwenStreamHandler implements StreamedAsyncHandler<String> {
    private CountDownLatch countDownLatch;

    public QwenStreamHandler(CountDownLatch countDownLatch){
        this.countDownLatch = countDownLatch;
    }

    @Override
    public State onStream(Publisher publisher) {
        publisher.subscribe(new Subscriber() {
            @Override
            public void onSubscribe(Subscription subscription) {
                subscription.request(Long.MAX_VALUE);
            }

            @Override
            public void onNext(Object obj) {
                try{
                    if(obj instanceof EagerResponseBodyPart){
                        EagerResponseBodyPart part = (EagerResponseBodyPart)obj;
                        byte[] bytes = part.getBodyPartBytes();
                        String words = new String(bytes,"UTF-8");
                        log.info(words);
                    }
                }catch(Throwable e){
                    log.error("系统异常",e);
                }
            }

            @Override
            public void onError(Throwable throwable) {
                log.error("系统异常",throwable);
            }

            @Override
            public void onComplete() {
                countDownLatch.countDown();
 
            }
        });
        return State.CONTINUE;
    }

    @Override
    public State onStatusReceived(HttpResponseStatus responseStatus) throws Exception {
        log.info("onStatusReceived: {}",responseStatus.getStatusCode());
        return responseStatus.getStatusCode() == 200 ? State.CONTINUE : State.ABORT;
    }

    @Override
    public State onHeadersReceived(HttpHeaders headers) throws Exception {
        return State.CONTINUE;
    }

    @Override
    public State onBodyPartReceived(HttpResponseBodyPart bodyPart) throws Exception {
        return State.CONTINUE;
    }

    @Override
    public void onThrowable(Throwable t) {
        log.error("onThrowable", t);
    }

    @Override
    public String onCompleted() throws Exception {
        return State.ABORT.name();
    }
}

maven依赖

java 复制代码 
<dependency>
    <groupId>org.apache.httpcomponents</groupId>
    <artifactId>httpclient</artifactId>
    <version>4.5.12</version>
</dependency>

<dependency>
    <groupId>org.asynchttpclient</groupId>
    <artifactId>async-http-client</artifactId>
    <version>2.12.3</version>
</dependency>

五、附带说明

5.1. 需要根据实际情况修改url，userId以及secret的信息

5.2. 如果AI服务已经实现了负载均衡，那么在URL中应该使用SLB（全局负载均衡）的IP地址

相关推荐
GISer_Jing2 小时前
神经网络初学总结（一）
人工智能·深度学习·神经网络
数据分析能量站2 小时前
神经网络-AlexNet
人工智能·深度学习·神经网络
Ven%2 小时前
如何修改pip全局缓存位置和全局安装包存放路径
人工智能·python·深度学习·缓存·自然语言处理·pip
YangJZ_ByteMaster3 小时前
EndtoEnd Object Detection with Transformers
人工智能·深度学习·目标检测·计算机视觉
volcanical4 小时前
Bert各种变体——RoBERTA/ALBERT/DistillBert
人工智能·深度学习·bert
知来者逆4 小时前
Binoculars——分析证实大语言模型生成文本的检测和引用量按学科和国家明确显示了使用偏差的多样性和对内容类型的影响
人工智能·深度学习·语言模型·自然语言处理·llm·大语言模型
跟德姆(dom)一起学AI4 小时前
0基础跟德姆（dom）一起学AI 自然语言处理05-文本特征处理
人工智能·python·深度学习·自然语言处理
云空5 小时前
《解锁分类神经网络预训练模型的奇妙世界》
人工智能·深度学习·神经网络
啊哈哈哈哈哈啊哈哈5 小时前
P7——pytorch马铃薯病害识别
人工智能·深度学习·学习
大知闲闲哟8 小时前
深度学习J6周 ResNeXt-50实战解析
人工智能·深度学习
热门推荐
0119个Web前端交互式3D JavaScript框架和库02ARM学习（31）编译器对overlay方式的支持03玄机平台应急响应—webshell查杀04RAG 实践- Ollama+RagFlow 部署本地知识库05【一文读懂】NTN（非地面网络）技术介绍06(欧拉)openEuler系统添加网卡文件配置流程、(欧拉)openEuler系统手动配置ipv6地址流程、(欧拉)openEuler系统网络管理说明07Ubuntu 20.04使用Livox mid 360 测试 FAST_LIO08组基轨迹建模 GBTM的介绍与实现（Stata 或 R）09Docker 夺命连环 15 问10大模型国产化适配7-华为昇腾LLM落地可选解决方案（MindFormers、ModelLink、MindIE）