Springboot接入OpenAi/Chat GPT的三种方式

由于现在网上的相关教程并不多外加没有使用代理的demo，所以抛砖引玉，写了三种调用方式，分别是直接访问、非官方SDK访问、官方开源的SDK访问

准备工作

1、导入pom文件（2023.3.30最新版本）

<dependency><groupId>com.unfbx</groupId><artifactId>chatgpt-java</artifactId><version>1.0.8</version></dependency><dependency><groupId>com.theokanning.openai-gpt3-java</groupId><artifactId>service</artifactId><version>0.11.1</version></dependency>

2、入参
这里的TalkDto 还可以入参maxToken，用于返回最大量的数据，官方计算token大小的方式：中文1字=3token 英文4字=1token。例如：哈 = 3；ha=1；哈ha=4；哈ha哈h=8
官方计算token地址：https://platform.openai.com/tokenizer

/**
 * @author panx
 * @date 2023/3/29 13:52
 */publicclassTalkDto{@ApiModelProperty("内容")privateString content;publicStringgetContent(){return content;}publicvoidsetContent(String content){this.content = content;}}

正文

1、直接访问

优点自定义且简便，具有高度的可玩性，所有openAi新增的入参都可以直接用。
缺点过于硬核，官方文档为纯英文。已经有轮子了为什么还要自行造车轮呢？
正文

由于我自行封装了请求方式，所以不展示代理怎么进行插入，主要展示Body层，这里是最简单的Body入参。

Map<String,Object> bodyMap =newHashMap<>();
        bodyMap.put("model","gpt-3.5-turbo");
        bodyMap.put("max_tokens","4096");{List<Map<String,String>> maps =newArrayList<>();//        Map<String, String> systemMap = new HashMap<>();//        systemMap.put("role", "system");//        systemMap.put("content", "你是一个分析师");//        maps.add(systemMap);Map<String,String> contentMap =newHashMap<>();
            contentMap.put("role","user");
            contentMap.put("content", talkDto.getContent());
            maps.add(contentMap);
            bodyMap.put("messages", maps);}Map<String,String> headerMap =newHashMap<>();
        headerMap.put("Content-Type","application/json");
        headerMap.put("Authorization","Bearer 123456789");

出参可以参考官方文档（https://platform.openai.com/docs/api-reference/chat）或下面的TalkResultDto

2、使用com.unfbx.chatgpt-java

项目地址

https://github.com/Grt1228/chatgpt-java

优点

纯中文文档，网上的demo也很多，更新速度也快，针对中国玩家的使用更便捷。

缺点

不支持别的类型的请求？粗略的找了一下没有找到。
有一些默认值无法去除，而不去除某一些值在长段入参中可能会报错，如果需要去除可能要等更新或自行去https://platform.openai.com/playground调试模型。

正文

@ApiOperation("非官方聊天api捏")@PostMapping("/talkByService")publicReTurnDto<String>talkByService(@RequestBodyTalkDto talkDto){//国内访问需要做代理，国外服务器不需要Proxy proxy =newProxy(Proxy.Type.HTTP,newInetSocketAddress("192.168.111.111",1080));HttpLoggingInterceptor httpLoggingInterceptor =newHttpLoggingInterceptor(newOpenAILogger());
        httpLoggingInterceptor.setLevel(HttpLoggingInterceptor.Level.BODY);OkHttpClient okHttpClient =newOkHttpClient.Builder().proxy(proxy)//自定义代理.addInterceptor(httpLoggingInterceptor)//自定义日志输出.addInterceptor(newOpenAiResponseInterceptor())//自定义返回值拦截.connectTimeout(10,TimeUnit.SECONDS)//自定义超时时间.writeTimeout(3000,TimeUnit.SECONDS)//自定义超时时间.readTimeout(3000,TimeUnit.SECONDS)//自定义超时时间.build();//构建客户端OpenAiClient openAiClient =OpenAiClient.builder()//构建.apiKey(Collections.singletonList("1234577987645"))//token.okHttpClient(okHttpClient)//注入okhttp.build();Message message =Message.builder().role(Message.Role.USER).content(talkDto.getContent()).build();ChatCompletion.ChatCompletionBuilder builder =ChatCompletion.builder();//聊天模型：gpt-3.5
        builder.model("gpt-3.5-turbo");
        builder.messages(Arrays.asList(message));ChatCompletion chatCompletion = builder.build();ChatCompletionResponse chatCompletionResponse = openAiClient.chatCompletion(chatCompletion);returnnewReTurnDto<>(chatCompletionResponse.getChoices().get(0).getMessage().getContent());}

3、使用com.theokanning.openai-gpt3-java.service

项目地址

https://github.com/TheoKanning/openai-java

优点

官方SDK。各种类型的请求都有单独的实体类，使用起来非常便捷。

缺点

更新速度没有上面的快，官方文档不太清晰。由于使用了 Retrofit+OkHttp 中国玩家需要进行繁琐的配置才能使用。

正文

@ApiOperation("官方api聊天捏")@PostMapping("/talkByOpen")publicReTurnDto<String>talkByOpen(@RequestBodyTalkDto talkDto){ObjectMapper mapper =newObjectMapper();
        mapper.configure(DeserializationFeature.FAIL_ON_UNKNOWN_PROPERTIES,false);
        mapper.setSerializationInclusion(JsonInclude.Include.NON_NULL);
        mapper.setPropertyNamingStrategy(PropertyNamingStrategy.SNAKE_CASE);//国内访问需要做代理，国外服务器不需要Proxy proxy =newProxy(Proxy.Type.HTTP,newInetSocketAddress("127.0.0.1",1549));HttpLoggingInterceptor httpLoggingInterceptor =newHttpLoggingInterceptor(newOpenAILogger());
        httpLoggingInterceptor.setLevel(HttpLoggingInterceptor.Level.BODY);OkHttpClient client =newOkHttpClient.Builder().proxy(proxy)//自定义代理.addInterceptor(newHttpLoggingInterceptor()).addInterceptor(httpLoggingInterceptor)//自定义日志输出.addInterceptor(newAuthOpenAiInterceptor("123456789000"))//token.connectTimeout(10,TimeUnit.SECONDS)//自定义超时时间.writeTimeout(3000,TimeUnit.SECONDS)//自定义超时时间.readTimeout(3000,TimeUnit.SECONDS)//自定义超时时间.build();Retrofit retrofit =newRetrofit.Builder()//构建者.client(client).baseUrl("https://api.openai.com/")//url.addConverterFactory(JacksonConverterFactory.create(mapper))//jackson.addCallAdapterFactory(RxJava2CallAdapterFactory.create())//.build();OpenAiApi api = retrofit.create(OpenAiApi.class);OpenAiService service =newOpenAiService(api);//text-davinci-003/* CompletionRequest completionRequest = CompletionRequest.builder()
                .prompt(talkDto.getContent())
                .model("text-davinci-003")
                .maxTokens(2060)
                .build();
        CompletionResult completion = service.createCompletion(completionRequest);
        return new ResponseVo<>(completion.getChoices().get(0).getText());*///gpt-3.5-turboChatCompletionRequest chatCompletionRequest =newChatCompletionRequest();ChatMessage chatMessage =newChatMessage();
        chatMessage.setRole(ChatMessageRole.USER.value());
        chatMessage.setContent(talkDto.getContent());
        chatCompletionRequest.setModel("gpt-3.5-turbo");
        chatCompletionRequest.setMessages(Collections.singletonList(chatMessage));ChatCompletionResult chatCompletion = service.createChatCompletion(chatCompletionRequest);returnnewReTurnDto<>(chatCompletion.getChoices().get(0).getMessage().getContent());}

后期工作

AuthOpenAiInterceptor 如下

importokhttp3.Interceptor;importokhttp3.Request;importokhttp3.Response;importjava.io.IOException;/**
 * @author panx
 * @date 2023/3/30 14:22
 */publicclassAuthOpenAiInterceptorimplementsInterceptor{privatefinalString token;publicAuthOpenAiInterceptor(String token){this.token = token;}publicResponseintercept(Interceptor.Chain chain)throwsIOException{Request request = chain.request().newBuilder().header("Authorization","Bearer "+this.token).build();return chain.proceed(request);}}

TalkResultDto 如下

/**
 * @author panx
 * @date 2023/3/29 14:35
 */publicclassTalkResultDto{privateString id;privateString object;privateInteger created;privateString model;privateUsageDto usage;privateList<ChoicesDto> choices;publicStringgetId(){return id;}publicvoidsetId(String id){this.id = id;}publicStringgetObject(){return object;}publicvoidsetObject(String object){this.object = object;}publicIntegergetCreated(){return created;}publicvoidsetCreated(Integer created){this.created = created;}publicStringgetModel(){return model;}publicvoidsetModel(String model){this.model = model;}publicUsageDtogetUsage(){return usage;}publicvoidsetUsage(UsageDto usage){this.usage = usage;}publicList<ChoicesDto>getChoices(){return choices;}publicvoidsetChoices(List<ChoicesDto> choices){this.choices = choices;}publicstaticclassUsageDto{privateInteger prompt_tokens;privateInteger completion_tokens;privateInteger total_tokens;publicIntegergetPrompt_tokens(){return prompt_tokens;}publicvoidsetPrompt_tokens(Integer prompt_tokens){this.prompt_tokens = prompt_tokens;}publicIntegergetCompletion_tokens(){return completion_tokens;}publicvoidsetCompletion_tokens(Integer completion_tokens){this.completion_tokens = completion_tokens;}publicIntegergetTotal_tokens(){return total_tokens;}publicvoidsetTotal_tokens(Integer total_tokens){this.total_tokens = total_tokens;}}publicstaticclassChoicesDto{privateMessageDto message;privateString finish_reason;privateInteger index;publicMessageDtogetMessage(){return message;}publicvoidsetMessage(MessageDto message){this.message = message;}publicStringgetFinish_reason(){return finish_reason;}publicvoidsetFinish_reason(String finish_reason){this.finish_reason = finish_reason;}publicIntegergetIndex(){return index;}publicvoidsetIndex(Integer index){this.index = index;}publicstaticclassMessageDto{privateString role;privateString content;publicStringgetRole(){return role;}publicvoidsetRole(String role){this.role = role;}publicStringgetContent(){return content;}publicvoidsetContent(String content){this.content = content;}}}}

如果没有代理的话，可以去购买一个香港的服务器，将服务部署在香港服务器中

畅爽使用~

标签： spring boot java 后端

本文转载自: https://blog.csdn.net/weixin_43184171/article/details/129858081
版权归原作者 紫薯千层糕 所有，如有侵权，请联系我们删除。

Springboot接入OpenAi/Chat GPT的三种方式

准备工作

正文

后期工作

畅爽使用~

发表评论

“Springboot接入OpenAi/Chat GPT的三种方式”的评论:

关于作者

overfit同步小助手

相关阅读

文章导航